Compare providers
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#
- Select the graph node that should supply the shared history.
- Enter one prompt in the composer.
- Select Send all.
- Watch the sibling responses stream and appear under the same parent prompt.
- Select each result node to inspect the answer, provider/model label, citations, and files.

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 the preferred response node and send the next prompt normally. That creates a descendant of the chosen provider answer. The other responses remain as siblings for reference; they are not included in the new active path.
Send all runs behind the current view, so you can continue inspecting the graph while responses finish.
Send all and synthesize#
Eligible accounts can choose Send All and Synthesize to run the four fixed provider routes—OpenAI, Gemini, Claude, and Grok—and then combine their completed responses. All four provider nodes and the pending synthesis node appear immediately. Focus moves to the synthesis node, where Convography shows each provider's progress while waiting for all four lanes to finish or stop.
The automatic router chooses the model, reasoning level, and whether web search is useful independently for each provider lane. A lane can therefore research current information before producing its response. The final synthesis does not perform another web search: it combines only the four completed provider responses and the conversation context supplied to it.
The completed response nodes are the only selected endpoints. Their shared branch history is included once, and the final assistant-only synthesis node has one incoming edge from each included provider response. It is positioned below and horizontally centered across its sources.
The original composer prompt becomes the fusion instruction automatically, so the final plain-text response directly answers the same request using all four provider branches. The synthesis streams into an ordinary assistant node; there is no structured-result wrapper or separate source-coverage panel.
You can select other graph nodes while the workflow runs. Stop on a provider node stops only that provider response, and that lane is omitted from synthesis. Once every provider lane is finished or stopped, synthesis proceeds from the completed responses when at least two remain. Stop on the pending synthesis node cancels the whole workflow. If the synthesis stage fails, every completed provider response remains available. A failed status can be dismissed immediately and otherwise disappears after a few seconds. Starting again creates a fresh four-response workflow; Convography never substitutes an older response.