Type anything.
It becomes a flow of real, runnable steps.
Pick an intent to see its real operation path. Every backend function is listed under the intent it serves; the link takes you to the exact place you use it. Functions marked agent are reached by your own agent through the MCP address on Account.
These verbs are in the design and not in the working menus. Nothing here runs; nothing is faked. They return to Start or the Workspace node menu when a real operation serves them.
Map not loaded yet.
Can this be answered? is part of Answer. Research with sources is part of Research: it runs a solve in research mode (retrieval forced, sources kept); the app has no separate retrieval-plan operation. Optimize a saved Flow against your stated objective: target → variable → score before/after → kept or rejected, with the reason.Your result lands here.
Ask one question. A completed solve shows its answer and cost, with its steps when the engine returns them.
Following the question through.
What runs: Reduce (break the question into steps) → Ground (check it against real facts) → Answer (write the answer) → Certify (check the answer before showing it)
Checking the cost and trace.
The first part of the answer arrived; the finished result has not. If this screen does not move on, the solve stopped early: ask the question again.
The free solve did not finish.
Your question is still here. Check your connection, then use “Try free solve again.”
If you are signed in, open Workspace and ask again. Otherwise, create an account or sign in first.
Worked result
The answer returned by a completed free solve appears here.
What ran, in order
Steps, in order
Blockers
Assumptions
Sources
Options by score
| Option | Score |
|---|
Your receipt · keep these IDs; they identify this answer and its checks.