How knightsky-cpu/col-workspace works
Agent Col is a browser-based persistent AI collaboration workspace. Users authenticate, choose a workspace, and submit chat turns; the FastAPI entry point delegates turn lifecycle and routing to the turn service, which coordinates responder and bounded specialist work and returns inspectable receipts. Separate workspace surfaces govern memory, notes, continuity, jobs, and artifacts, with Firestore-backed persistence and optional Google speech/model integrations. The README documents these capabilities; unsampled implementation paths are represented only where the tree supports them, and edges avoid asserting unshown internal wiring.
Main entry points: agent_col_turn_service.py Turn orchestration · main.py FastAPI routes · database.py Firestore data · app.mjs Workspace UI
Components 23
Browser workspace
- Workspace UI
frontend/app.mjs - API client
frontend/api.mjs - Sign-in context
frontend/auth-view.mjs
API and orchestration
- FastAPI routes
main.py - Authentication
auth.py - Turn orchestration
agent_col_turn_service.py - Turn routing
agent_col_routing.py - Responder runtime
supervisor_runtime.py - Agent jobs
agent_job_repository.py
Governed collaboration
- Profile memory
memory_proposals.py - Collaborative notes
collaborative_note_service.py - Session continuity
continuity_service.py - Working state
working_state_service.py - Preference learning
preference_learning_service.py
Specialist work
- Research specialist
research_expert_service.py - Source specialist
source_expert_service.py - Computation specialist
computational_expert_service.py - Requirements review
requirements_verification_service.py - Artifact lifecycle
generic_artifact_service.py
Persistence and integrations
- Firestore data
database.py - Cloud speech
speech_service.py
Other components
- User
- Google Cloud services
Connections 29
- User → Workspace UI: uses
- Workspace UI → Sign-in context: builds sign-in context
- Workspace UI → API client: requests APIs
- API client → FastAPI routes: calls same-origin APIs
- FastAPI routes → Authentication: verifies identity
- FastAPI routes → Turn orchestration: executes chat turn
- Turn orchestration → Turn routing: requests routing
- Turn orchestration → Responder runtime: runs response
- Turn orchestration → Research specialist: delegates work
- Turn orchestration → Source specialist: delegates work
- Turn orchestration → Computation specialist: delegates work
- Turn orchestration → Requirements review: delegates work
- Turn orchestration → Artifact lifecycle: queues artifacts
- FastAPI routes → Profile memory: serves decisions
- FastAPI routes → Collaborative notes: serves note actions
- FastAPI routes → Session continuity: serves choices
- Turn orchestration → Working state: uses context
- Turn orchestration → Preference learning: records feedback
- FastAPI routes → Agent jobs: manages jobs
- FastAPI routes → Firestore data: persists workspace data
- Turn orchestration → Firestore data: reads and writes turns
- Agent jobs → Firestore data: stores job records
- Profile memory → Firestore data: stores memory state
- Collaborative notes → Firestore data: stores note state
- Artifact lifecycle → Firestore data: stores artifacts
- FastAPI routes → Cloud speech: handles speech routes
- Cloud speech → Google Cloud services: uses speech services
- Responder runtime → Google Cloud services: uses model runtime
- Research specialist → Google Cloud services: grounds research