**Send behind** grows the graph without moving the chat pane to the new result. It is useful when you want another answer to run while you keep reading, writing, or working on the current node.

## Use the main composer

1. Select the graph node that should be the parent.
2. Enter a prompt in the composer.
3. Choose **Send behind** instead of **Send**.

The request uses the selected path, but the current view stays put. When the response completes, its node is available in the graph and may receive an unread indicator.

The desktop shortcut is **Ctrl+Shift+Enter**. On mobile, open the send-mode menu and choose **Send behind**.

![The Send behind control creating an unread child while the original node stays open](/images/send_behind.webp)

## Use the inline send-behind control

Assistant messages expose a small send-behind action. Select it to open a floating composer anchored to that answer, enter the follow-up, and send. This is the safest way to target a specific older answer in a long visible path because the parent is explicit.

Inline send behind is not shown for a pending answer or a context-only message.

## When it is most useful

- launch several scenario checks from the same answer;
- request a shorter, more technical, or more skeptical version in the background;
- keep taking notes on one branch while another provider responds;
- add a research sub-question without losing your reading position.

## What does not change

Send behind does not merge answers, make the background result part of the active path, or change the selected node. To continue from the result, select its node after it appears.

The action is disabled while **Load context** is active, while a file clarification is waiting, or when messaging is unavailable because of access or usage state.

## Send all versus Send behind

**Send all** is a comparison action: it sends one prompt behind the selected point to the available major provider routes and creates sibling responses. **Send behind** creates one response using the current Auto, provider, or model selection.


Grow a branch in the background while keeping the current node open.

Graph & branches

Send behind

Meet Convography and learn why conversations are organized as graphs.

Get started

Convography is designed for work that does not move in a straight line. A normal chat hides every abandoned alternative in the scrollback. Convography keeps those alternatives visible as a graph, so you can investigate, compare, and return without starting over. What makes Convography different The main workspace is a conversation graph . Each node represents a point in the conversation, and each connecting line shows which earlier point gave rise to it. Click a node to open its full ancestor path in the chat pane. Send a message from there to continue that path or create a new branch. This structure gives you three useful freedoms: Explore without overwriting. Revisit an earlier decision and try another direction while the original remains intact. Compare in context. Send the same prompt to multiple providers and inspect the resulting sibling nodes. Synthesize deliberately. Select the completed text branches that matter and combine their full paths into one connected response. Reuse the useful part. Split a selected path into a new independent conversation or share only the branch another person needs. A simple conversation graph branching from one question into two alternatives How a conversation grows A new conversation starts with an empty graph and an open composer. Your first prompt and its answer establish the first node. A normal follow up continues after the selected node. If you select an older node and send, Convography preserves its existing descendants and adds another branch. Only the path leading to the selected node is shown in the docked chat pane. That path is also the conversational history used for the next request, together with any files or context you deliberately include. A neighboring branch does not silently leak into the active path. What you 

Introduction

Sign up, accept the beta terms, and recover access when needed.

An account keeps your conversations, graph structure, files, shared links, usage mode, and subscription state together. Sign up and account recovery are available from the authentication screens. Create an account 1. Select Start free on the Convography website. 2. Enter the requested account details and choose a password. 3. Complete any email verification shown for your environment. 4. Review and accept the beta welcome terms before entering the workspace. 5. Choose Auto or Simple when the setup screen asks how you want to use models. Auto is the suggested starting point. You can change the usage mode later under Account settings → Models & Providers . The sign up and welcome flow from account creation to usage mode selection Sign in Use the email address and password attached to the account. After authentication, Convography restores your conversation list and opens the workspace. If a long idle session can no longer be refreshed, the app asks you to retry or sign in again instead of treating the stale session as current. Reset a forgotten password 1. Open the sign in page and choose the password recovery option. 2. Enter your account email. 3. Enter the verification code delivered by email. 4. Choose a new password and confirm it. Passwords created through account settings or recovery must be at least 12 characters. Recovery codes and reset tokens are time bound; request a new code if the current one expires. If an email does not arrive, check the address carefully, inspect spam or junk folders, wait a few minutes, and then retry once. Contact support@convography.com if delivery continues to fail. Accept the beta welcome The beta welcome appears after authentication until its terms are accepted. It explains current access and records acceptance on the account. You c

Create your account

Understand the conversation list, graph canvas, and docked chat pane.

The workspace has three coordinated surfaces: the conversation pane, the graph canvas, and the chat pane. The graph is the primary surface; the other two help you choose a body of work and inspect one path through it. Conversation pane The left conversation pane is your library. Use it to: create a new conversation; select an existing conversation; rename a conversation; delete an entire conversation after confirmation; collapse the pane when you want more canvas space. Conversation titles are generated from the work and can be renamed. The list is ordered by recent activity. Graph canvas The center canvas shows the selected conversation as connected nodes. Drag an empty part of the canvas to pan and scroll or pinch to zoom. Selecting a node marks it active and opens its path in the chat pane. Lines show parent to child relationships; pastel accents make branch families easier to follow. The graph remains interactive while the chat pane is open. Convography shifts the graph into the visible space beside the pane so the selected node does not disappear underneath it. The three workspace areas labeled Conversation pane, Graph canvas, and Chat pane Docked chat pane The chat pane opens from the left side of the canvas when you select a node. It shows the sequence of messages from the root to that node—not every message elsewhere in the graph. From the pane you can: send a normal follow up; send behind the current node; load context without requesting a reply; compare major providers with Send all ; run Send All and Synthesize on eligible plans; select files and file scope; choose model controls in Simple mode; inspect citations, included files, response metadata, and exports. Close the pane with the X to return to an unobstructed graph. Selecting another node opens or immed

Tour the workspace

Create a conversation, send the first prompt, and follow the new path.

Your first conversation establishes the root of a new graph. You do not need to design the graph in advance—start with the question or outcome you care about, then branch when an alternative becomes useful. Create the conversation 1. Select New conversation in the conversation pane. Convography immediately switches to the new empty conversation; you do not need to select it again from the list. 2. If the chat pane is not open, choose Write the first message on the canvas. 3. Enter a clear first prompt in the composer. 4. In Auto mode, simply select Send . In Simple mode, optionally choose a provider or model, reasoning level, and supported tool first. 5. Watch the response stream into the chat pane and the new node appear in the graph. An empty graph becoming the first conversation node after a prompt is sent Write a useful opening prompt Give the model the goal, important constraints, and desired form of the answer. For example: text Help me plan a launch brief for a privacy focused note taking app. Identify the audience, the core promise, and three positioning risks. Use a short table followed by recommendations. You can attach files before sending when source material matters. In Simple mode, use web search only when the answer needs current public information. Continue the active path After the answer finishes, the response node becomes the active point when the selection has not changed during generation. Type a follow up and select Send to extend it. The graph adds the next connected node and the pane follows it. If you click elsewhere while a response is streaming, Convography does not force the selection back when the response completes. The finished node remains available in the graph and may be marked as unread. Try a first branch Select the first answer again

Start your first conversation

Learn the vocabulary behind nodes, paths, branches, context, and splits.

Convography uses a small set of concepts consistently across the graph, chat pane, files, and sharing. Conversation A conversation is one saved graph in the conversation pane. It has a title, a selected model or provider route when applicable, connected message nodes, and related files. Deleting a conversation removes the entire graph after confirmation. Node A node is a point in the graph. A normal exchange is presented as one readable graph unit even though the application preserves the user prompt and assistant response needed to reconstruct the path. Context only entries and file relationships can also appear in the graph. Parent, child, and path The parent is the earlier node a new node follows. A child is a direct continuation from that point. The active path is the chain from the root through the selected node. That path is what the docked chat pane displays. Branch A branch forms when a node has more than one continuation. Branching does not duplicate or erase the original path. Select an earlier node and send a new prompt to add an alternative child. A labeled graph showing a root, parent, child, active path, and two branches Active node The active node is the point currently selected in the graph. It controls which path the chat pane displays and where a normal send attaches. Selection has meaning—check the highlight before sending when branch placement matters. Send behind Send behind adds a new continuation from a chosen point without moving the active view to the result. Use it to launch background alternatives and keep reading or composing where you are. Context only node Load context stores information on the graph without asking a model for a response. Later descendants on that path can use the stored information. It is useful for facts, constraints, or 

Core concepts

Read the conversation graph and understand its visual structure.

The graph is a visual map of a conversation. Read it from earlier nodes toward later nodes: connecting lines show which point each continuation came from, and sibling nodes reveal alternatives created from the same parent. Read the structure Each message node includes a concise generated title or label so you can scan the idea without opening the full transcript. The selected node has a stronger visual state. Unread completed responses can carry an unread marker when you moved elsewhere while they were generated. Pastel branch accents distinguish major branch families. The first line of work uses the main green accent; additional branch roots receive other colors. A color follows that branch through its descendants unless you apply an override from the node menu. A graph annotated with the root, sibling branches, selected node, unread response, and branch colors Understand direction Convography lays nodes out by depth. A node farther from the root represents a later point along a path. Multiple children at the same depth are alternatives, not a sequence that the model reads together. Ordinary conversation paths are tree shaped: a normal node has one conversational parent, so its chat ancestor path is unambiguous. A synthesis result is the exception in the graph. It can have incoming edges from several explicitly selected source nodes while its own chat pane remains an assistant only starting point. Files can appear as related file nodes and connect to the message nodes that used them; those relationships do not change a normal conversational parent chain. Select a node Click a message node to select it. The docked chat pane opens the complete ancestor path and scrolls to the source prompt associated with that graph point. Selecting another node updates the pane immediat

Graph overview

Pan, zoom, select, recenter, and move between conversation paths.

Graph navigation is independent from chat scrolling. You can move around the canvas while the selected path stays open, then choose a different node when you want the chat pane to change. Pan the canvas Drag an empty area of the graph to move the canvas. Start the drag outside a node so you do not change the selection. When graph navigation mode is active, panning keeps keyboard focus on the current node so you can continue with the Arrow keys immediately. Panning is useful when a graph has several distant branches or when the docked chat pane reduces the visible canvas width. Zoom and fit Use a mouse wheel or trackpad gesture to zoom. On touch screens, pinch with two fingers. When graph navigation mode is active, press + or to zoom in or out. Press 0 or double click empty canvas to run the same fit visible operation. Empty canvas fitting preserves keyboard focus on the current node. Press F to frame the selected node and its ancestor path. Outside synthesis selection mode, double clicking a conversation node runs that fit active branch operation for the node you clicked. During synthesis selection mode, double clicking a node toggles its synthesis checkbox instead. Double clicking another control does not fit the graph. The standalone ? button in the bottom right corner of the graph opens the keyboard shortcut reference. You can also press ? while a graph node has focus. The graph chooses an initial fit that keeps the content inside the available canvas. On desktop, opening a chat card from the closed state keeps the current zoom and recenters the selected node in the canvas area beside the card. Selecting another node while the card is already open, or closing the card, does not automatically fit or recenter. Mobile keeps its existing separate Graph to Chat transition

Navigate the graph

Find ranked lexical matches across conversation titles, tags, turns, summaries, and decisions.

Use the search button immediately beside the conversation panel expand or collapse control to find text across your conversations. On mobile, use the equivalent search shortcut. You can also press Ctrl+K or Command+K while focus is outside an editable field. Opening search places keyboard focus in the query field so you can type immediately. Conversation search showing the query business, scope controls, and highlighted matching results Choose the search scope Search starts with All conversations . Switch to This conversation to restrict results to the active graph. Results are grouped by conversation for scanning while retaining their overall relevance order. Convography searches conversation and node titles, stored tags, user and assistant text, stored summaries, and structured decision metadata. It does not search profile information, file names, or text inside uploaded files. Write a lexical query Search uses PostgreSQL full text matching, not embeddings or semantic similarity. It matches words and their language stems; it does not infer unstated synonyms or meaning. Put a phrase in quotes to require adjacent words, such as "pricing model" . Use OR between alternatives, such as postgres OR firestore . Prefix a word with to exclude it, such as database sqlite . A stored tag may appear as a matched field, but a tag is simply a short metadata label attached to a conversation or message by an existing workflow. Search does not add a tag editor. The specific queries How did we get here? , Show the path here , and Show context for this node return the selected node's ancestor path. Select a graph node first; without one, search asks for an anchor. Other question shaped queries are ordinary lexical searches. Read and open results Results are ordered by relevance. Strong ti

Search conversations

See how a selected node determines the visible history and model context.

Selecting a node does more than highlight a card. It defines the conversation path shown in chat and the historical context for the next message. What the chat pane includes The pane reconstructs the ancestor chain from the conversation root to the selected node. It shows the user and assistant messages on that route, plus context only entries, citations, attachments, and available response metadata. Messages on sibling branches are not part of this view. This prevents an alternative you explored elsewhere from silently influencing the selected line of thought. One highlighted active path through a graph while neighboring branches remain dimmed Where a normal send attaches When a node is selected, Send attaches the new request after that node. If it already has another child, the new request becomes another branch. If it is a leaf, the graph simply grows forward. After generation, Convography normally selects the new response and advances the chat pane. If you changed graph selection while the response was running, the app respects your newer selection instead. Combined graph nodes and chat messages The graph favors readable units, while the chat pane shows the precise message sequence. A graph node can summarize the paired prompt and response with separate generated titles behind the scenes. This is why selecting one graph point can reveal both the source prompt and its answer in the path. A synthesis node is assistant only. Although the graph displays an incoming edge from every source selected for synthesis, opening the result does not choose one of those source branches or replay them above the response. Its linear chat path starts with the synthesized answer. A normal follow up continues from that answer. Context boundaries The active ancestor path is the core cont

Understand nodes & paths

Return to any earlier node and explore a different direction without losing work.

Create a branch when you want a different continuation from an earlier point without deleting or rewriting the path you already have. Branch from any node 1. Find the point before the conversation took the direction you want to change. 2. Select that node in the graph. 3. Confirm the docked chat pane ends at the intended prompt or answer. 4. Write the alternative follow up in the composer. 5. Select Send . If that node already has a continuation, the new response appears as a sibling branch. Convography preserves both paths and gives the new branch its own visual accent. Selecting an earlier node and sending a new prompt to create a sibling branch Good reasons to branch test a different assumption or constraint; ask for a different audience, tone, or output format; pursue a counterargument without losing the primary case; try another model after the same background; separate “what if” scenarios in planning; compare a concise answer with a deep analysis. Make the branch easy to understand later Start with a prompt that names the difference. “Try again” produces a weak label and makes branches hard to distinguish. Prefer “Alternative: assume the budget is cut by 40%” or “Challenge the security model from an attacker’s perspective.” Use the node menu to assign a branch color when the graph contains several related scenarios. Color is visual organization only; it does not change the model context. Branch versus Send behind Both actions create a continuation from a graph point. A normal Send follows the result and usually makes it active. Send behind creates the result while keeping your current view in place. Use Send behind when you want to launch alternatives and inspect them later. Branch versus Split conversation A branch stays inside the current graph and shares its ea

Create a branch

Send behind grows the graph without moving the chat pane to the new result. It is useful when you want another answer to run while you keep reading, writing, or working on the current node. Use the main composer 1. Select the graph node that should be the parent. 2. Enter a prompt in the composer. 3. Choose Send behind instead of Send . The request uses the selected path, but the current view stays put. When the response completes, its node is available in the graph and may receive an unread indicator. The desktop shortcut is Ctrl+Shift+Enter . On mobile, open the send mode menu and choose Send behind . The Send behind control creating an unread child while the original node stays open Use the inline send behind control Assistant messages expose a small send behind action. Select it to open a floating composer anchored to that answer, enter the follow up, and send. This is the safest way to target a specific older answer in a long visible path because the parent is explicit. Inline send behind is not shown for a pending answer or a context only message. When it is most useful launch several scenario checks from the same answer; request a shorter, more technical, or more skeptical version in the background; keep taking notes on one branch while another provider responds; add a research sub question without losing your reading position. What does not change Send behind does not merge answers, make the background result part of the active path, or change the selected node. To continue from the result, select its node after it appears. The action is disabled while Load context is active, while a file clarification is waiting, or when messaging is unavailable because of access or usage state. Send all versus Send behind Send all is a comparison action: it sends one prompt be

Turn a passage in an answer into a focused follow-up branch.

Use Ask about this to turn a specific passage in an assistant answer into a focused follow up. The resulting request is anchored behind that answer, so the branch keeps the path that produced the passage. Ask about a passage 1. Open the path containing the assistant answer. 2. Select the exact text you want to discuss. 3. Choose Ask about this from the selection menu. 4. Add your instruction in the floating composer. 5. Send the question. Selecting text in an answer to reveal the Ask about this action The Ask about this composer opened for the selected passage The selected passage is carried into the follow up affordance so you do not need to copy and paste it manually. The request grows from the answer that contains the selection. Write a focused instruction The selection supplies the subject, but the model still needs a task. Useful additions include: “Explain this in plain language.” “What evidence supports this claim?” “Rewrite this for an executive audience.” “Identify the hidden assumption.” “Show a concrete example.” Selection boundaries The action is intended for assistant text in the rendered response. Selecting controls, file metadata, or content outside the message does not create the same anchored prompt. Very long selections can also produce an unnecessarily broad request; choose the smallest passage that preserves the question. How it differs from copying Copying a passage into the main composer can work, but the branch attaches wherever the graph is currently selected. Ask about this associates the follow up with the answer you selected, reducing the risk of sending from the wrong node. After sending, the new result can remain in the background like an inline Send behind request. Select its graph node when you are ready to inspect or continue it.

Ask about selected text

Store a context-only node when you want to add facts without generating an answer.

Load context adds information to a path without asking a model to answer immediately. It creates a context only node that later descendants can inherit. Store context 1. Select the node after which the new information should apply. 2. Turn on Load context in the composer actions. 3. Enter the fact, constraint, excerpt, or instruction. 4. Select Store context . The graph adds the context point, but no assistant generation or provider charge is initiated for an answer. The stored message appears with context only styling in the chat path. A context only node inserted between an answer and later follow up Good uses record a decision: “The launch date is now November 14.” add a constraint: “Do not recommend tools that require customer data export.” preserve a source note before asking multiple questions; attach a file and explain why it matters; add meeting feedback without interrupting the graph with a generated response. Context follows the path The entry affects descendants on the same path. It does not become global memory for every conversation or retroactively change sibling branches that diverged before it. Branch after the context node when every new alternative should share the information; branch before it when you want a scenario that excludes it. Interactions with other send modes While Load context is active, Send behind and Send all are disabled because those actions request model responses. Turn Load context off to return to normal generation. On mobile, the send button label changes to reflect the active mode, and the long press send menu lets you switch among normal send, Send behind, and Load context. Context versus files A context only note is text in the ancestor path. A selected file is separate source material included according to file scope and provi

Load context without a reply

Send one prompt to major providers and inspect sibling responses in the graph.

Use Send all to ask the same prompt through the available major provider routes. Convography creates one shared user prompt and multiple sibling response nodes, making the comparison visible in the graph. Run a comparison 1. Select the graph node that should supply the shared history. 2. Enter one prompt in the composer. 3. Select Send all . 4. Watch the sibling responses stream and appear under the same parent prompt. 5. Select each result node to inspect the answer, provider/model label, citations, and files. One prompt fanning out to OpenAI, Gemini, Claude, and Grok response nodes The comparison is limited to providers and models that are currently available, allowed for your account, and supported by the comparison route. The control names OpenAI, Gemini, Claude, and Grok, but an unavailable route can be omitted or fail independently. Partial failures One provider failure does not discard successful responses from the others. A failed graph node records its failure state so the comparison remains structurally honest. Open a failed result to read the available error and retry later or use a different route. Compare fairly Keep the prompt, selected ancestor path, and file selection the same. Provider capabilities still differ: a model may support different reasoning controls, hosted tools, file handling, or output limits. Treat the comparison as evidence about the complete route—not just the brand name. A useful review checklist is: Dimension What to inspect Accuracy Claims, calculations, and cited evidence Coverage Important constraints or cases omitted Clarity Structure, explanations, and audience fit Traceability File citations, web sources, and uncertainty Efficiency Whether the answer reaches a useful result without unnecessary work Continue from a winner Select 

Compare providers

Select text branches and combine their shared context into one multi-parent response.

Use Synthesize to combine several points in a conversation graph into one new response. Synthesis is available to eligible Trial, Pro, Max, and administrative accounts; it is not available on Core. Select source nodes Open synthesis selection mode from the graph controls. The selection bar appears at the bottom of the graph, and a temporary checkbox appears on every graph node. Ordinary text responses, stored text context nodes, and earlier text synthesis nodes can be selected. Image, file, artifact, pending, failed, and other unsupported nodes remain available for inspection but cannot be checked. A normal click still opens that node in the chat pane. Use this to read a branch before deciding whether to include it. Checking the node's box selects it without changing the open chat. On desktop, double clicking the node is a selection shortcut. Checkboxes disappear when selection mode ends. Choose at least two nodes. There is no selection count limit, and the order in which you check nodes does not affect the result. Use the instruction box at the top of the selection panel to describe the result you want—for example, ask for a decision memo, a comparison, or a consolidated implementation plan. It starts with Synthesize these conversations. You can replace that text, but the instructions cannot be blank. The selected node count and actions appear beneath the box. Six selected graph nodes and custom fusion instructions ready to synthesize Choosing Synthesize immediately closes selection mode, removes the checkboxes and selection bar, and returns the graph to normal navigation. The source nodes remain fully opaque and visually unchanged while the pending synthesis node is translucent. You can open any other node, inspect its chat path, use graph menus, and continue working 

Synthesize graph nodes

Copy a selected path into a new independent conversation.

Split conversation copies the path ending at a selected node into a new independent conversation. Use it when a branch has become its own project rather than another alternative inside the same graph. Create the split 1. Select the final node you want included. 2. Open the node menu. 3. Choose Split conversation . 4. Convography creates a new conversation, switches to it, and selects the copied endpoint. The new conversation appears in the conversation pane with its own identity and title. Only the ancestor path leading to the selected node is copied; unrelated sibling branches are not included. Selecting a branch to split from the original conversation The selected branch copied into its own conversation What becomes independent After the split: new messages in the copy do not appear in the original; deletion in one conversation does not delete the other; each conversation can be renamed separately; each conversation has separate sharing links; the copied path remains available even if the source branch is later removed. Split versus share a branch Split when you need a separate editable workspace. Share a branch when someone else needs a read only view of one path. A signed in viewer may create their own branch copy from a share when that share permits it, but that action creates work in the viewer’s account. Split versus create a branch Keep alternatives in one graph while they benefit from side by side visual comparison. Split when the graph’s topic, audience, owner, or lifecycle has diverged enough that separate organization is clearer. Naming the new conversation The application derives a safe title from the copied work. Rename it from the conversation pane if the new purpose is more specific. A good split title names the outcome, such as “Launch brief — enterpris

Split a conversation

Use titles, branch colors, selection, and deliberate structure as a graph grows.

Large graphs stay useful when branches have clear purposes and visible boundaries. Convography provides generated titles and branch colors; your prompts and branching choices provide the structure. Branch at meaningful decisions Create branches where assumptions, audiences, scenarios, or methods diverge. Avoid branching every minor follow up. A graph with five meaningful scenario roots is easier to read than one with twenty nearly identical retries. Name the difference in the first prompt on a branch. Generated node titles can then reflect the distinction. Use branch colors deliberately Open a node menu and choose a color to override the automatic branch accent. The chosen color follows that branch visually. Use Auto color to return it to the default palette. One simple scheme is: green for the main or recommended path; pink for risk and counterargument; purple for research; yellow for open questions; blue for implementation. Colors do not change context or priority. Document your scheme in an early context node if other people will use the graph. A large graph organized into colored research, risk, and recommendation branches Combine mature branches After exploring alternatives, use Synthesize to select the text or context endpoints you want to combine. Convography includes each selected endpoint's complete ancestor path, includes shared history once, and creates one assistant only result with an incoming edge from every selected endpoint. Ordinary follow up messages still use only their single active ancestor path; sibling branches are included only when you explicitly select them for synthesis. Split mature work When one branch develops its own audience or long sequence, use Split conversation . This keeps the source graph readable while preserving the complete path 

Organize complex graphs

Understand deletion boundaries before removing a node or branch.

Deleting a graph node starts a reversible deletion. Convography immediately hides the selected conversational turn and every reply below it, returns selection to the parent when possible, and waits seven seconds before permanent cleanup. Delete a non root node 1. Select the node you intend to remove. 2. Use the toolbar trash action or open the node menu and choose Delete . 3. Read the confirmation. It names the node and warns that replies below it are included. 4. Confirm only after checking the descendants on the graph. If the selected visual node represents an assistant answer paired with its user prompt, deletion starts at the corresponding user turn so the graph is not left with a broken half exchange. A deletion confirmation highlighting the selected node and all descendants that will be removed Root deletion Deleting the root is equivalent to deleting the conversation. The confirmation asks whether to remove the conversation and all of its messages. This removes it from the conversation pane. Related files When messages are deleted, Convography evaluates files related to that subtree. Files that are no longer referenced can be cleaned up with the deleted work. Do not assume every exported or generated artifact will survive a subtree deletion; download anything you must keep first. Synthesis sources and active workflows Deleting a synthesis node removes that result and its ordinary descendants but leaves its selected source branches in place. Deleting a source node is different: the affected set also includes synthesis results that depend on that source and any descendants below those results. Read the confirmation and graph carefully because a source can influence a result outside its ordinary parent child subtree. If the deleted area contains optimistic provider 

Delete nodes safely

Switch between Chat and Graph views and navigate branches on a touch screen.

On mobile, Convography separates the chat path and graph canvas into focused views. The underlying conversation structure is the same as desktop. Open the graph From Chat, tap Graph . The graph view includes controls for the conversation pane, a new conversation, sharing, deletion, feedback, account settings, and returning to Chat. Pan with one finger on empty canvas space and pinch with two fingers to zoom. A single tap selects a node for graph actions. Double tap a message node to open its path and return to the chat experience. The Convography Chat view on mobile The Convography Graph view on mobile Use the conversation pane Open the conversation shortcut from the graph controls to reveal the conversation list. Select a conversation or create a new one, then close the pane to use the canvas. The sidebar backdrop closes the panel without changing your selection. Choose a send mode The mobile composer keeps the primary send action compact. Long press or open the send mode menu to choose: Send — create the next response and follow it; Send behind — create a background continuation without changing your view; Load context — store a context only message; Send All and Synthesize — when available, create four provider responses and combine the completed lanes after every lane finishes or stops; at least two completed responses are required. The send label reflects the active mode. Send all remains a separate comparison action when it is available. Avoid accidental branch placement Mobile screens show less of the graph at once, so confirm the selected node before composing. Double tap it, read the final prompt in Chat, and then send. If you need to keep reading the current path while launching another, use the inline send behind action on the exact assistant answer. Improve 

Use the graph on mobile

Use the composer, stop a response, and understand where a new reply lands.

Chat & models

The composer always sends from a specific place in the graph. Before writing, verify the selected node and the end of the visible chat path—especially when a conversation has several branches. Send a normal message 1. Select the graph node you want to continue, or open a new empty conversation. 2. Type in Send a message… . 3. Add files or Simple mode controls if needed. 4. Select Send or press Enter . Use Shift+Enter to add a new line without sending. A normal send creates a new child after the selected point and follows the completed response unless you changed selection while it was running. The composer with Files, model controls, send modes, and Send labeled Watch a response arrive Convography adds an optimistic graph node as soon as the request begins and streams answer tokens into the chat pane. A work log can appear before or alongside the final answer when the route provides safe progress or reasoning summary events. You can move to another node while generation continues. The response still finishes on its original parent. The app does not pull you back to it; look for the new or unread node in the graph. Stop generation Select Stop while the current response is streaming. Convography cancels the browser stream and removes the temporary pending state. Provider work that already completed upstream cannot always be reversed, so stopping is a control over the active request experience rather than a guarantee that no usage occurred. Choose another send behavior The same composer supports several graph operations: Action Result Send Generate one response and follow it Send behind Generate one response without moving the current view Send all Generate sibling responses through available major provider routes Send All and Synthesize Generate four provider responses, t

Write & send messages

Let Convography choose a model, reasoning level, and supported tool for each prompt.

Auto mode keeps model, reasoning, and hosted tool controls out of the workspace. Convography analyzes the task and chooses an allowed route for the prompt. What Auto chooses Depending on the prompt and currently available catalog, Auto can choose: a provider and concrete model; a reasoning effort appropriate to the task; web search for questions that need current public information; a lower cost route for simple classification, extraction, or summarization; a stronger route for complex research, mathematics, or system design. The server owns the final policy. Auto never grants access to a disabled model or bypasses the account’s plan, usage limit, provider availability, or tool restrictions. A prompt entering Auto and routing to a model, reasoning level, and optional web search Use Auto effectively Describe the task, source material, and freshness requirement explicitly. Compare: text Summarize the following paragraph in three bullets: … with: text Research the current regulatory status of this proposal and cite public sources. The first supplies text and asks for a bounded transformation. The second explicitly needs current research. Clear framing helps Auto distinguish source words inside supplied content from instructions about the outside world. See what Auto used Completed assistant messages show the actual model and, when available, reasoning effort and enabled hosted tools. This metadata remains attached after reload so you can understand how the answer was produced. For a multi provider comparison, use Send all rather than trying to ask Auto for four independent answers in one response. Switch to Auto 1. Open Account settings . 2. Choose Models & Providers . 3. Select Auto under Usage mode. 4. Return to the app. The model picker, reasoning control, and Web toggl

Use Auto mode

Choose a provider or model and reveal guided reasoning and tool controls.

Simple mode exposes guided model controls while preserving server side safety and availability rules. You can select Auto, a provider route, or a concrete mainline model from the composer. Turn on Simple mode 1. Open Account settings → Models & Providers . 2. Select Simple under Usage mode. 3. Return to the workspace. The composer now shows the model menu. Supported reasoning and Web controls appear only for selections that offer them. The Simple mode model menu with Auto, provider routes, and model submenus Choose a route The model menu can contain: Auto — use the global router for this conversation; a provider route — let Convography choose a suitable model within OpenAI, Gemini, Claude, or Grok; a specific model — use a visible, allowed model from that provider’s submenu. The available list is dynamic. A model can be hidden, disabled, or temporarily unavailable based on the live catalog, provider policy, plan, validation state, or current rollout. Provider route versus exact model A provider route keeps you within one provider but allows a model choice that matches the task. An exact model gives greater repeatability for comparisons. Prefer the provider route for everyday use; choose a model when its behavior or capabilities are specifically relevant. Selection persists with the conversation The active model choice is stored on the conversation where appropriate. Check the composer after switching conversations, because another conversation can have a different model or provider route. Controls remain request specific Reasoning effort and Web are chosen for the next message and supported route. The answer records what was actually used. If a model does not support a control, the control is absent or disabled; Convography does not simulate support in the browser. Mana

Use Simple mode

Choose supported reasoning effort and web search settings per message.

In Simple mode, the composer shows reasoning and hosted tool controls only when the selected route supports them. These settings apply to the request you are about to send. Choose reasoning effort Open the reasoning menu and choose from the presets exposed for the selected model. Possible labels include No reasoning, Minimal, Low, Medium, High, Extra high, and Max, but no model is required to offer every level. Use lower effort for direct rewriting, extraction, classification, or simple questions. Use higher effort for multi constraint planning, difficult analysis, proofs, architecture, or tasks where checking intermediate implications matters. The reasoning preset menu and Web Search toggle in the composer Higher reasoning is not automatically more accurate. It can take longer and use more of the plan’s AI allowance. A clear prompt and relevant source material usually matter more than selecting the maximum level. Turn on Web Search Select Web to request the supported provider’s hosted web search. Use it when the answer depends on current public information, such as recent events, present product details, or an up to date policy. When search returns citations, the answer includes a Web Sources section with titles, domains, and links. Open the sources and verify that each one supports the nearby claim. Request an image Select Image in Auto mode to route the next request to an available image model. The route treats the request as an image generation or image editing instruction even when the wording is conversational or not explicitly visual. For example, What is 2+2? produces a visual answer instead of being handled as an ordinary text question. Know the boundaries Web and Image are request scoped composer routes and cannot be active together. Tool availability varies b

Reasoning & hosted tools

Follow live output, stop generation, and read safe progress summaries.

Responses stream into the chat pane as they are generated. Convography also preserves a safe work log when the provider or route supplies progress steps or a reasoning summary. Streaming states After you send, the graph adds a pending node and the chat pane shows an active response. Text appears incrementally rather than waiting for the complete answer. A caret marks the live edge of the response. When generation finishes, Convography stores the final content, model metadata, citations, attachments, usage details, and completion status. Reloading the conversation shows the persisted answer rather than replaying the stream. A streaming answer with active work log steps followed by the final response Read the work log The work log is labeled Thinking and can include steps marked active, done, or failed plus a concise reasoning summary. It is useful for seeing whether a request is researching, processing files, generating an image, or preparing an answer. The log is not a transcript of private model reasoning. It contains only safe, supported summaries and progress events. Some routes provide no log, in which case the pane shows a normal loading state until answer text arrives. Stop a response Select Stop to abort the active browser request. The pending graph state is cleaned up and you can revise the prompt. Depending on timing, upstream work and associated usage may already have occurred. Navigate while work continues You can select another node or conversation while a response streams. Convography tracks each active generation by its original tree and parent. When a background response completes, it remains attached to the correct branch. It does not replace your newer selection. Failed responses A provider or persistence error can produce a failed response state. Succe

Streaming & the work log

Read file citations, web sources, and source metadata attached to an answer.

Convography can display two kinds of answer evidence: citations to uploaded file chunks and links returned by hosted web search. They have different provenance and should be reviewed differently. File sources Answers grounded in processed files can include a Sources section. Each citation identifies: a source number used within the answer context; the original file name; a chunk position and shortened chunk ID; a retrieved text snippet; a relevance score when available. Use the file name and snippet to confirm that retrieval selected the intended material. A high relevance score describes matching behavior, not factual authority. An answer showing file Sources, Web Sources, and an Included Files popover Web sources When Web Search is used, Web Sources lists linked page titles and domains. Open each relevant result in a new tab and check publication date, authorship, and direct support for the claim. Web citations are tied to the answer that used search. They are not a permanent guarantee that the page will remain available or unchanged. Included Files The files button beneath a message shows Included Files for that request. This is the clearest record of which selected files were actually carried into the turn. It is distinct from generated attachments displayed with an answer. If the app asks which file you mean, choose from the clarification options. This happens when a prompt refers ambiguously to “the file” while several candidates are active. Cite responsibly 1. Open the cited source. 2. Verify the exact claim in context. 3. Prefer primary material for important decisions. 4. Check dates for time sensitive information. 5. Treat uncited model statements as unverified synthesis. A citation shows where supporting context came from; it does not certify that the model i

Sources & citations

Copy an answer or export a response as PDF, Markdown, text, or HTML.

Every completed assistant response has actions for reuse. Available actions depend on whether the answer contains text, generated images, attached files, or a pending/failed state. Copy text Use the copy action beneath an answer to copy its rendered source text. Code blocks also have their own copy control so you can copy only the code without surrounding explanation. Review copied text before pasting it into another system. Markdown markers, citations, or math delimiters can render differently outside Convography. Export a response For a completed text response, open Download response and choose: PDF document ( .pdf ) Markdown ( .md ) Plain text ( .txt ) HTML document ( .html ) Convography generates a file artifact, adds it to the file library and graph relationships, and starts the download. The exported artifact reflects that response, not the entire conversation graph. The Download response menu with PDF, Markdown, text, and HTML options Download images and attachments Generated images have a direct download action and can also be selected from their preview. Other message attachments expose their file name as a download control when a signed URL is available. Download URLs are short lived. If an older link has expired, use the action in Convography again to request a fresh URL. Find exports later Generated exports are stored as app created files. Open Account settings → Files to browse or download them. They can also appear as file nodes connected to the message that produced them. What is not exported A single response export does not automatically include sibling branches, hidden provider internals, account details, share tokens, or the complete graph topology. For now, export important answers individually. Full conversation and bulk graph export are planned sep

Copy, download & export

Work with rendered Markdown, code blocks, tables, and LaTeX output.

Assistant answers render GitHub flavored Markdown with additional math support. Formatting is part of the response display; it does not change the underlying graph structure. Markdown and tables Headings, emphasis, lists, links, block quotes, task lists, and tables render in the chat pane. Wide tables stay within the pane and can scroll rather than stretching the workspace. If a model returns malformed Markdown, copy the answer or ask it to reformat the content. Convography does not silently rewrite the semantic content to repair every formatting mistake. Code blocks Fenced code renders in a contained code block with its own copy action. Long lines and large blocks stay within the chat surface. Always review code before running it, especially shell commands, migrations, or scripts that alter files or external systems. python def branch label(topic: str, scenario: str) str: return f"{topic} — {scenario}" Inline code uses a compact monospace style and is kept distinct from mathematical delimiters. Math Convography renders LaTeX style inline and display math, including common model output using \(...\) and \[...\] delimiters. For example, an answer can render an inline expression such as \(E = mc^2\) or a display equation: text \[ P(A \mid B) = \frac{P(B \mid A)P(A)}{P(B)} \] Long display equations remain contained within the chat pane. Code spans and fenced code blocks are protected from math normalization so programming syntax is not treated as an equation. A response containing a Markdown table, copied code block, and rendered display equation Images in Markdown Images referenced by a response render inside a bounded frame and can open in a focused tab. Generated images are also represented as attachments with download and grounding details. Never treat an image link fr

Code, tables & math

Browse uploaded, generated, and exported files from account settings.

Files & images

The file library collects files you uploaded and files Convography created, including generated images and exported responses. Open it from Account settings → Files . Browse the library The Files tab shows the file name, source, size, added date, and available actions. Use the source filters to view: All files; Uploaded files; files Created by app . Search filters by file name. Use descriptive names before upload so a file is easy to recognize in the library and composer. The account file library with search, source filters, selection, and file rows Understand file states An upload becomes usable only after storage, checksum, MIME type, and security processing succeed. A file can be listed but temporarily unavailable for AI context if processing is incomplete or its type is unsupported by the selected provider. The chat file picker disables unavailable files and explains that they are not ready or cannot be used for AI context yet. Refresh the library after a short wait if an upload is still processing. Select files for bulk actions Use each row’s checkbox or Select all visible files . The selection bar can download one file directly, download several as files.zip , or delete selected files. Bulk download and delete accept up to 100 files at a time. Filters affect what “select all visible” means. Clear the search and source filter before assuming every library file is selected. Storage and duplicates Uploads are stored as account owned canonical files. If identical content is uploaded again, Convography can reuse the existing canonical file instead of storing a duplicate. The file can still be selected for the current message. Library versus graph The library is account wide management. The graph shows how files relate to a particular conversation and its nodes. Deletin

Use the file library

Upload or select files from the composer and send them with a prompt.

Use the Files button in the composer to upload new material or select an existing account file for the next request. Upload from the composer 1. Open the path where the file should be used. 2. Select Files in the composer. 3. Choose Upload files . 4. Select one or more files from your device. 5. Wait for each file to become selectable. 6. Select the files you want and write the prompt. 7. Send normally, behind, or as a provider comparison. The Files button shows a count when context is selected. The composer file panel showing uploads, existing files, and selected context Select an existing file Search the file panel by name, then choose a ready tile. Selected tiles remain attached to the composer context until you change the selection or the path’s file state changes. Supported processing varies by provider and model. Common documents, PDFs, text, images, audio, video, and office formats can follow different direct processing paths. Spreadsheet AI processing can be unavailable even when the file itself is safely stored and downloadable. Trust the picker’s current enabled state rather than a generic extension list. Tell the model what to do Attaching a file does not define the task. Name the file and the expected operation: text Using “research notes.pdf,” list the three strongest claims and quote the supporting section for each. Separate direct evidence from your inference. When several files are active and “the file” is ambiguous, Convography asks you to choose the intended file before continuing. Upload failures An upload can fail because of a size or account storage limit, a checksum mismatch, content type validation, a security status, network interruption, or storage availability. Read the displayed reason. Rename alone will not make an unsupported or unsafe file 

Add files to a chat

Choose files from a node, branch, or full conversation and inspect what was included.

File context is branch aware. The composer lets you build a selection from files associated with the current node, the active ancestor path, or the entire conversation. Choose a scope shortcut In the Files panel, use: Node — select files uploaded or used at the current node; Branch — select files associated along the active path; Conversation — select available files from anywhere in the conversation. These controls update the explicit file selection. Review the selected count and tiles before sending; the scope label is not a permanent automatic policy for every future request. Node, Branch, and Conversation file scope controls over a branching graph Why branch scope matters Suppose one graph branch analyzes a confidential draft while a sibling analyzes a public brief. Using Branch limits the working set to the selected lineage. Conversation scope is broader and can mix material from alternatives, so use it only when the prompt truly needs cross graph files. Inspect what a message used After sending, a file count action beneath the message opens Included Files . It lists the exact file records carried into that turn with type, size, and date. Use this record when reviewing provenance or diagnosing why an answer referenced an unexpected document. Generated attachments shown on the answer are outputs, not necessarily inputs. File citations are retrieved source chunks, while Included Files records the request’s file context. File references can carry forward The active path preserves selected file snapshots and branch local file history. Generated images can also establish branch local reference lineage for later edits or continuity prompts. Visiting a sibling branch does not intentionally replace the selected branch’s reference history. Still, make important references e

Control file context

Open graph file nodes, download individual files or ZIPs, and remove files safely.

Files can be previewed from graph relationships and managed from the account library. Preview support depends on format and browser capability; download remains the fallback. Preview from the graph Select a connected file node or open the file list attached to a message node. Convography can render: images; audio and video with browser controls; PDF and safe document views; plain text and source files; Markdown with formatted headings, lists, code, and tables. For unsupported formats, the preview explains that it is unavailable and provides a download action when possible. A graph file node opening an in app preview panel with a Download action Some generated HTML, SVG, DOCX, or ODT artifacts are isolated or sandboxed for safer preview behavior. A preview is not a guarantee that active content is trustworthy. Download a file Use the file row action in Account settings → Files , the graph preview footer, or the attachment action under a message. Convography requests a fresh signed URL for owner authorized downloads. These URLs expire, so regenerate the download from the interface instead of bookmarking it. Select several files in the library to download a ZIP. Bulk actions are limited to 100 files. Delete files 1. Select one or more files in the library. 2. Choose Delete . 3. Read the warning that deleted files cannot be recovered and will no longer be available to conversations. 4. Confirm. Deletion removes the account owned file record and begins storage cleanup. It can affect past conversations that display or depend on the file. Export or download a copy first when retention matters. If a download fails Refresh the library and try the action again to obtain a new signed URL. If the file was deleted, quarantined, or partially cleaned up, it may no longer be recoverabl

Preview, download & delete files

Request images, understand grounding details, and download generated results.

Convography can route image generation and editing prompts through plan appropriate, available image models. Generated images are stored as file artifacts and attached to the response node. Request a new image Describe the subject, composition, medium, lighting, aspect or orientation, and important exclusions. If you need several distinct outputs, state the count clearly. text Create three editorial illustrations of a branching research process. Warm cream background, deep green line work, subtle pastel branch accents, landscape composition, no text or logos. Auto can infer an image task when the capability is available. In Simple mode, a visible image capable route can be selected directly. Available models and output options vary by Core, Pro, or Max tier and live provider status. A generated image response with model, quality, resolution, dimensions, and download action Edit or continue an image Attach or select the reference image and describe what should change and what must remain. Branch local image lineage helps continuity prompts resolve the most recent relevant image, including a named character or a clear request such as changing a background material. For reliable edits, be explicit: text Using the selected kitchen image, change only the upper cabinets from painted wood to walnut. Preserve the layout, camera angle, lighting, and every appliance. If no eligible image exists on the selected branch, Convography returns a missing reference error instead of editing an unrelated image. Understand result metadata Generated attachments can show provider, model, quality, requested resolution, actual pixel dimensions, reference image count, and grounding sources. Select the image or use Download image to save it. Grounded image requests Some supported Gemini image rou

Generate images

Create a public link for a whole graph or only the selected path.

Sharing

Sharing creates a time limited public snapshot. Choose the whole graph when another person needs the alternatives, or a selected branch when they need only one path. Create a link 1. Open the conversation you want to share. 2. To share one path, select its final node first. 3. Choose Share conversation in the graph toolbar, or open a node menu and choose Share branch . 4. Select Whole graph or Selected branch . 5. Choose Create graph link or Create branch link . 6. Convography copies the URL to the clipboard when browser permissions allow it. The Share Conversation dialog with Whole graph and Selected branch options What each scope includes Scope Shared structure Whole graph The complete conversation snapshot with all included branches Selected branch Only the ancestor path ending at the selected node Both scopes preserve a read only graph and path viewer. Images are included when sharing creates the snapshot, and signed in members are allowed to branch a copy into their own workspace under the current share settings. Snapshot behavior A share captures conversation messages at creation time. Work added to the owner’s graph later is not silently appended to an existing public snapshot. Create a new link when recipients need an updated version. The title stored for a share comes from the conversation or selected branch at creation. Your active links list can show the current source title for easier management, but the public snapshot remains its own record. Expiration Links expire seven days after creation. The share dialog displays each active link’s expiry. Create a fresh link for access beyond that time. Share safely Anyone with the URL can open the public snapshot until it expires or is revoked. Before creating it: inspect every included branch; remove secrets and unn

Share a graph or branch

Navigate a read-only shared graph and branch it into your own account when allowed.

A shared conversation opens in a public, read only viewer. Visitors can navigate the shared graph and inspect paths without access to the owner’s account or original workspace. Navigate shared work Select a graph node to open the path leading to it. Whole graph shares expose the captured alternatives; branch shares contain one ancestor path. The viewer displays answer content, web citations, included generated images, and failure or context only states that were preserved in the snapshot. A public shared conversation viewer with graph, selected path, and sign in actions The public copy intentionally omits owner only controls, private account metadata, uploaded source file access, and internal identifiers that are not needed for the viewer experience. Sign in or sign up Visitors can read without an account. The viewer offers Sign in and Sign up actions for people who want to create their own editable copy. Branch to your workspace When a signed in member views a link that allows branching: 1. Select the endpoint to copy. 2. Choose Branch to my workspace . 3. Convography copies the selected shared lineage into a new conversation owned by that member. 4. The member is redirected to the new conversation and can continue independently. Included shared images are copied into the new owner’s file storage when available. The new conversation does not stay synchronized with the source or public link. Unavailable links A viewer receives an unavailable message when the token is invalid, expired, revoked, or the snapshot can no longer be read. Ask the owner for a new link. Do not send the old token to anyone except trusted support through an approved channel. Evaluate shared content Sharing does not certify an answer. Check cited web pages, distinguish user prompts from model outpu

View shared work

Copy, rotate, revoke, and review active links from one place.

Manage public access from the Share dialog for one conversation or from Account settings → Shared links across the account. Review active links The Share dialog lists unrevoked, unexpired links for the open conversation. Account settings groups active links by source conversation. Each entry identifies its title, graph or branch scope, and expiration. Expired and revoked links are excluded from the active list. The Shared links account tab grouped by conversation with copy, rotate, and revoke controls Copy a link Use Copy link to put the active URL on the clipboard. If clipboard access is blocked, the Share dialog displays the URL in a field so you can copy it manually. Rotate a link Rotate when the URL may have reached someone unintended but you still want the snapshot to remain available to new recipients. Rotation invalidates the old token and returns a new URL for the same share record. Send the replacement only to the intended audience. Anyone using the prior link will no longer have access. Revoke a link Revoke when public access should end. Revocation is immediate and cannot be undone from the interface. Create a new share if access is needed later. Revoking the link does not delete the owner’s conversation or copies that signed in members already branched into their own workspaces. Use expiration as a backstop Every link expires after seven days, but expiration is not a substitute for revocation. Revoke promptly when a project changes audience, content is found to be sensitive, or a link is posted in the wrong place. Audit habits Review Shared links regularly: 1. Group links by conversation. 2. Confirm each recipient still needs access. 3. Revoke unnecessary links. 4. Rotate any link with uncertain distribution. 5. Create a fresh snapshot when the source work ha

Manage shared links

Update your profile, timezone, password, and session access.

Account & billing

Account settings collect personal profile, security, model, file, subscription, and sharing controls. Open them from the workspace settings menu. Update your profile The Profile tab lets you edit your full name and timezone. Your email address and verification state are displayed with the profile but are not edited through the same form. Choose the timezone that should be used for account facing dates and future localized behavior. Save the form and wait for the success notification before leaving the page. Account settings with the Profile and Security tabs highlighted Change your password 1. Open the Security tab. 2. Enter the current password. 3. Enter and confirm the new password. 4. Choose the update action. The new password must be at least 12 characters. Use a unique passphrase or password manager generated value. The form rejects mismatched confirmation before sending the request. If you no longer know the current password, sign out and use the password recovery flow from the sign in page. Sign out Use Sign out in Account settings or the workspace settings menu. Sign out before leaving a shared device. A session that cannot refresh after a long idle period returns you to a retry or sign in state rather than leaving unrestricted access. Replay the intro Account settings includes Replay intro tutorial . It prepares the tutorial conversation and returns to the workspace. Replaying does not delete your existing conversations. Security habits Do not share recovery codes, access tokens, share tokens, or provider credentials. Review Shared links and revoke public URLs that are no longer needed. Remove sensitive content before sending it to a provider. Use separate browser profiles on shared computers. Contact support@convography.com for suspicious account behavior. Acc

Profile & security

Choose a usage mode and understand app-managed provider availability.

The Models & Providers tab controls how much routing choice appears in the workspace and shows current provider/model availability. Choose a usage mode Two modes are available: Mode Workspace behavior Auto Hides model, reasoning, and hosted tool controls; the server chooses an allowed route Simple Shows guided provider/model selection plus supported reasoning and Web controls The selection is saved to your profile. Switching mode does not rewrite existing message metadata or change answers already generated. The Models and Providers tab with Auto and Simple usage mode cards App managed provider access Current plans use app managed access for OpenAI, Gemini, Claude, and Grok routes made available by the service. Provider cards show availability, but member entered API keys are reserved for future BYOK plans and are not part of the current setup workflow. Never send a provider key to support or paste one into chat because a provider card appears unavailable. Read the model catalog In Simple mode, the page displays visible current models grouped by provider or model family. Status can include Available, Validated, Not validated, or Unavailable, with a reason or last test information when applicable. Managed models can be sendable without member validation. Some image models are listed through image specific availability rather than a billable text validation request. For the current eligibility snapshot across Trial, Core, Pro, and Max, see Model availability by plan. The matrix is checked against application routing policy in continuous integration. Refresh status Choose Reset to refresh provider and model information. If the page says model access is temporarily unavailable: 1. wait briefly and refresh once; 2. check whether the workspace shows a plan or usage restrictio

Models & providers

Compare current app-managed providers and models across Trial, Core, Pro, and Max.

The model catalog is plan aware. Use this page to check which app managed provider routes each subscription plan is eligible to select in Simple mode. Auto mode uses the same account entitlements while choosing an appropriate route for the task. Eligibility does not guarantee continuous availability. A provider outage, temporary application restriction, usage limit, account state, or model retirement can make an eligible route unavailable. The model picker and response status in the app are the live source of truth. Provider access by plan All four current plan families use app managed access to OpenAI, Gemini, Claude, and Grok. Members do not need to add provider API keys. The beta Trial receives the same image model tier as Pro during its active trial period. Plan Provider access Included providers Trial App managed OpenAI, Gemini, Claude, Grok Core App managed OpenAI, Gemini, Claude, Grok Pro App managed OpenAI, Gemini, Claude, Grok Max App managed OpenAI, Gemini, Claude, Grok Model eligibility by plan A check means the model is included in that plan's current routing policy. A dash means it is not included. The list covers current selectable chat, research, and image generation routes; it does not list retired, hidden, or unsupported provider catalog entries. Provider Model Type Trial Core Pro Max OpenAI GPT 5 Nano gpt 5 nano Chat ✓ ✓ ✓ ✓ OpenAI GPT 5.6 Luna gpt 5.6 luna Chat ✓ ✓ ✓ ✓ OpenAI GPT 5.6 Terra gpt 5.6 terra Chat ✓ ✓ ✓ ✓ OpenAI GPT 5.6 Sol gpt 5.6 sol Chat ✓ ✓ ✓ ✓ OpenAI GPT Image 2 gpt image 2 Image ✓ ✓ ✓ ✓ OpenAI GPT Image 2.5 Flare gpt image 2.5 flare Image — — — ✓ OpenAI GPT Image 2.5 Sunburst gpt image 2.5 sunburst Image — — — ✓ Gemini Gemini 3.5 Flash Lite gemini 3.5 flash lite Chat ✓ ✓ ✓ ✓ Gemini Gemini 3.6 Flash gemini 3.6 flash Chat ✓ ✓ ✓ ✓ Gemini

Model availability by plan

Understand trial access, included AI usage, warnings, and rolling limits.

Convography uses plan based AI credits to account for provider work. Plans define included capacity and image generation access, while the application records the actual provider usage it can measure. Current plan families The catalog contains a 30 Day Beta Trial with the full Pro experience and three paid plan families: Core for lighter everyday work; Pro for frequent research, creation, and larger projects; Max for intensive workflows and the highest image capacity. During the current soft launch, Core and Max are shown as coming soon and paid checkout is not enabled for new purchases. Use the live Subscriptions page as the source of truth for availability and pricing. See Model availability by plan for the current provider and model eligibility attached to each plan family. The Subscriptions page with trial, Core, Pro, and Max plan cards How credits work AI credits are an application accounting unit, not provider tokens. A request can involve text tokens, generated images, search, retries, title generation, or other provider calls. Convography normalizes recorded cost into the account’s included allowance. Per message token counts, provider costs, and charged credit calculations are not displayed in the chat pane; use the workspace warnings and Subscriptions page for account level usage status. Usage states The application can show: a low credit warning for the current billing period; Included usage exhausted , which blocks new AI work until capacity changes; a rolling Daily limit reached , which asks you to try later; plan access or payment states that restrict messaging. Warnings appear in the workspace and Subscriptions page. A daily boundary blocks new work; it does not necessarily cancel provider work already processing. Manage consumption Use Auto for task appr

Plans & AI credits

Review plan state and use the Stripe portal when paid billing is available.

The Subscriptions tab shows plan and access state. Existing paid members can open Stripe’s hosted billing portal when a Stripe customer subscription is attached to the account. Review subscription state Plan cards can show current plan, trial state, coming soon availability, a scheduled change, past due status, blocked status, or cancellation timing. The surrounding message explains what action is currently possible. Subscription status and the Open Stripe billing portal action Open the Stripe portal Choose Open Stripe billing portal to manage an existing paid subscription. The action is unavailable until the account has started a paid subscription. Portal redirects are created server side; do not trust a billing link sent through an unrelated message or chat response. Plan changes When paid checkout is enabled: an upgrade keeps the same Stripe subscription and charges the remaining price difference according to the current policy; the higher AI allowance appears after Stripe confirms the invoice; a downgrade is scheduled for the next billing cycle; the current cycle allowance remains until that boundary; cancellation timing is displayed on the active plan card. During the present soft launch, new paid checkout is disabled even though plan cards and supporting billing flows are implemented. Do not infer purchase availability from a displayed price alone. Payment problems A past due plan directs you to update the payment method in Stripe. A blocked plan must be reactivated before AI access returns. If Stripe shows success but Convography does not update after a reasonable delay, avoid making a second payment and contact support with the account email and approximate time—never full card details. Billing safety Start billing changes from Convography Account settings. Conf

Manage billing

Replay the intro and use contextual tips to learn in the workspace.

Help & reference

Convography teaches the graph progressively: a guided intro explains the workspace, while contextual tips appear around relevant actions as you use them. Intro tutorial The intro uses a prepared conversation to demonstrate: the conversation pane and conversation management; the graph canvas, pan, zoom, branches, and node selection; the docked chat pane and composer controls; sharing a whole graph or current branch; the settings menu and where to continue learning. Tutorial copy adapts to the mobile layout, including double tap node behavior and the separate Chat and Graph views. The intro tutorial spotlighting the conversation graph and explaining branches Start or replay the intro The intro starts after initial setup when model access is ready. To replay it later: 1. Open the workspace settings menu and choose the replay action, or open Account settings. 2. Choose Replay intro tutorial . 3. Wait while Convography prepares the tutorial conversation. 4. Follow the highlighted steps in the workspace. Replaying creates or restores tutorial state; it does not remove ordinary conversations. Dismiss and resume You can exit the tutorial when you need the normal workspace. Progress is stored so the experience does not behave like an untracked overlay. Replay it from settings when you want to start the guided sequence again. Contextual tips Tips are short, dismissible explanations triggered by eligible product actions. They cover advanced behavior after the main tutorial, such as branch operations, sharing, or composer features. The system records which tips were shown or dismissed so the same hint does not interrupt every session. Tips are assistance, not modal requirements. Dismiss one after you understand it and use the documentation for the complete workflow. If the tutorial

Tutorials & tips

Use documented keyboard controls and accessible interaction patterns.

Convography provides keyboard behavior for the composer, graph, menus, dialogs, and documentation. Graph traversal can update the selected chat path while keeping keyboard focus in the graph until you explicitly enter the chat pane. Composer shortcuts Shortcut Action Enter Send the current message Shift+Enter Insert a new line Ctrl+Shift+Enter Send behind without switching nodes Escape Dismiss an inline send behind composer On a Mac, the Send behind implementation accepts the Command modifier in the same combined shortcut. Documentation shortcuts Shortcut Action Ctrl+K or Command+K Open documentation search / Open search when focus is not in an input Escape Close search or mobile documentation navigation Graph access Opening a node moves focus into its chat pane. Press Alt+G ( Option+G on macOS) from anywhere outside an open dialog to center and focus the selected graph node. If the selected node is not visible, the shortcut centers and focuses the conversation root. On mobile it opens the Graph view first. This starts graph navigation mode: chat and route updates do not take focus from the destination node, and only Enter or Space hands focus into the chat pane. Graph nodes are excluded from the normal Tab sequence. Once a graph node has focus, use: Shortcut Action Alt+G or Option+G Center and focus the selected graph node, or the conversation root as a fallback Tab or Shift+Tab Stay on the current graph node; focus does not move Arrow Up Select the visible parent and update the chat pane; do nothing at a root Arrow Down Select the first visible child and update the chat pane; do nothing at a leaf Arrow Left Select the previous sibling and update the chat pane; stop at the first sibling Arrow Right Select the next sibling and update the chat pane; stop at the last sibl

Keyboard & accessibility

Resolve common session, model, file, streaming, sharing, and usage issues.

Start with the message shown in the interface. Convography distinguishes session, provider, file, usage, sharing, and storage failures so the next action can be specific. The workspace stays on Loading session 1. Wait for the bounded refresh attempt to finish. 2. Use the displayed retry action once. 3. If refresh fails, sign in again. 4. Disable an aggressive privacy extension for the site only if it blocks authentication cookies. 5. Contact support if the branded loading or recovery state never advances. Do not clear all browser data before checking whether unsaved composer text matters. No models are available Open Account settings → Models & Providers and choose Reset once. Check for an account plan or usage restriction. Current plans use app managed access; adding a personal key is not the remedy. Provider quota or billing failures can temporarily block an entire provider’s validation batch. A response failed or stopped Open the failed node and read its error. Check whether sibling Send all responses succeeded. Retry later for a provider outage. Reduce large or unsupported file context. Choose another available provider in Simple mode. Check the Subscriptions tab for exhausted or daily limited usage. Avoid rapid retries. Accepted calls may still consume provider usage. A synthesis did not complete For direct synthesis, confirm that at least two completed text or text context nodes are selected and that the fusion instructions are not blank. For Send All and Synthesize , open the workflow nodes and check each provider status. Fusion needs at least two completed provider responses after every lane becomes terminal. A provider lane Stop omits only that lane. Stop on the pending synthesis node cancels the complete compound workflow. A failed direct synthesis restores it

Troubleshooting

Understand practical account, sharing, provider, and deletion boundaries.

Convography stores account and conversation data to provide graph history, files, sharing, and billing. Your own handling choices—especially file uploads and public links—remain an important part of privacy. Conversation data Prompts, responses, graph relationships, selected file context, model metadata, citations, and supported usage records are persisted so a conversation can be reloaded accurately. Only the active ancestor path is sent as conversational history for a normal branch request; sibling branch content is not silently merged into it. Provider processing AI requests are sent to the provider selected by Auto, the provider route, or the chosen model. Files included for direct processing can also be sent to that provider. Remove secrets and data that the provider does not need. Read the current Privacy Policy and Terms of Service for the governing commitments. This user guide explains product behavior but does not replace those documents. Files Uploaded and app created files are account owned records with bounded signed download URLs. Security and processing states determine whether a file can be used. Deleting a file makes it unavailable to conversations and starts storage cleanup; deletion is not reversible from the interface. Public sharing A share link is a bearer URL: anyone who has it can open the snapshot until expiration or revocation. Whole graph shares can expose more than the selected path. Use branch scope, inspect content before sharing, and revoke links whose audience is uncertain. Signed in members can copy an allowed shared path into their own account. Revoking your public link does not erase copies another member already created. Billing and telemetry The application records provider usage and billing account state needed for credit accounting,

Privacy & data

Quick answers about graphs, branches, context, models, files, sharing, and billing.

What is a conversation graph? It is a tree shaped map of a conversation. Nodes are conversational points, edges show parent to child relationships, and multiple children from one node are alternative branches. Does a new branch include sibling branches as context? No. The selected node’s ancestor path is the core conversational history. Siblings remain visible in the graph but are not silently included. How do I create a branch? Select an earlier node, confirm its path in chat, and send a new follow up. If it already has a continuation, the new result becomes another child branch. What is Send behind? It sends from a chosen point but keeps your current view and selection in place. Use it for background alternatives. The desktop shortcut is Ctrl+Shift+Enter. What is Load context? It stores a text entry on the path without asking a model to reply. Descendants of that context node can use the information. What is the difference between a branch and a split? A branch stays inside the same graph. Split conversation copies the selected ancestor path into a separate, independent conversation. How do I compare models? Use Send all to create sibling responses from the available major provider routes. Or use Simple mode to choose a specific provider or model for controlled branch by branch comparisons. Does Auto show which model it used? Yes. The completed assistant message records the actual model and, when available, reasoning effort and enabled hosted tools. Can I use my own provider key? Not on current app managed plans. Provider key entry is reserved for future BYOK plans. Which file types are supported? Storage accepts a range of files, but AI processing support varies by provider and format. The Files picker is the live authority: selectable files are ready for the active 

Send behind

Use the main composer#

Use the inline send-behind control#

When it is most useful#

What does not change#

Send all versus Send behind#