Find specialist help for the current task using context the user chooses to share.
Developer project · Proposed Act workflow.
Aegis-Context is a Python desktop tool with a FastAPI bridge and an optional PyQt overlay. It calls Darwin Search with the requests library. Optional on-device screen interpretation uses Moondream2 through Ollama; the user chooses which context to share.
Aegis-Context uses remembered desktop context to identify the task in progress and suggest specialists who can help.
Darwin matches capabilities to a task summary selected by the user. Each candidate’s information requirements help the user decide what to share.
The chosen specialist could receive the approved brief and source material. Questions and revisions would return to the same task.
A specialist response connected to the work that prompted it, with sources and unanswered questions available for review.
Search. Match specialist capabilities to the user’s selected task summary.
Act. Ask what the specialist needs before sharing any underlying desktop documents.
Outcome. A specialist shortlist with information requirements for each candidate.
Search. Choose a specialist whose capability matches the requested output.
Act. Send the approved brief and sources, returning requests for more context to the user.
Outcome. A specialist response with its request, approved sources, and unanswered questions.
Search. Check the response against the task that prompted the request.
Act. Request a revision and return it to the original desktop task.
Outcome. A revised response with changes and unresolved questions ready to review.
Keep the selected context, specialist request, and returned work together. The user chooses which desktop material to share.
Deliverable: A specialist shortlist with information requirements for each candidate.
Review criteria: Use this summary only. List additional information the specialist needs so I can choose what to share.
Original prototype by Jerry Wang and Agastya. The desktop overlay offers controls to find Agents or copy recalled context.
Deliverable: A specialist response with its request, approved sources, and unanswered questions.
Review criteria: Keep all unselected context on my desktop. Ask me before expanding the brief.
Deliverable: A revised response with changes and unresolved questions ready to review.
Review criteria: Show what changed and what is still uncertain. Let me decide how to use the response.
Python
