Research¶

Use Research Mode when you want Colossus to investigate a question and leave a cited report you can inspect later. A research run plans bounded queries, gathers released evidence, extracts claims, and records limitations alongside its report. The run and its sources are durable; the terminal's current mode is not.

Ask a question in the terminal¶

For one question, enter it directly as a command:

/research How does Colossus recover an interrupted workflow?

This starts a research run in the current session without changing the terminal's mode. To ask several research questions in a row, enter Research Mode first:

/research on

Then submit an ordinary prompt, for example:

Which parts of this repository enforce approval before a tool effect?

While Research Mode is active, ordinary prompts go to the research service rather than the usual agent turn. Use /research status to check the current mode, and /research off to return to Execute Mode. The bare /research command toggles between Research and Execute modes. The mode resets when the TUI restarts.

Research Mode does not accept image attachments. If images are queued in the composer, use /detach all before submitting a research question.

Know what the terminal will search¶

The terminal route uses standard depth and selects repository, web, and configured MCP evidence lanes. Colossus bounds collection in each lane and records unavailable, denied, failed, or skipped attempts as limitations. Web and MCP collection require their own configured routes and permissions; the question cannot choose a provider or grant itself access.

Research produces a source-backed report, not just search results. Its stable source labels, such as R1, tie material claims to the evidence Colossus released. The final report is also appended to the session conversation.

Find the report and its evidence¶

Inside the TUI, list the research runs belonging to the current session:

/research list

Copy a run ID from the list. The CLI can show the full report, source labels, and extracted claims:

colossus research show RESEARCH_RUN_ID
colossus research sources RESEARCH_RUN_ID
colossus research claims RESEARCH_RUN_ID

/research status reports the terminal mode; research show reports the durable run status. When reviewing a report, read its limitations as well as its citations. A missing lane may leave a useful partial answer, but it does not count as evidence for a claim that depends on that lane.

If Colossus stops during research, the run is marked interrupted on recovery and is not retried automatically. Inspect the recorded run and any external effects before starting another question.

Choose a narrower scope from the CLI¶

Use a shell command when you need to choose depth or evidence lanes explicitly. For example, keep a repository investigation in the same session:

colossus --approval-mode ask research run \
  "How does effect authorization work?" \
  --session SESSION_ID --source repo --depth standard

The CLI accepts quick, standard, or deep depth and any selected combination of repo, web, and mcp sources. A run without --session creates a fresh session. See Deep research for lane setup, bounds, and the complete investigation workflow.

What's next?¶