Web UI
Web UI
Package: agentic-orchestration-web/
Role
Browser chat front-end using WebSockets. Each user message spawns the Python tool with dynamic flags:
python main.py --dynamic "<text>" --no-save --no-verify(and optional session), orpython main.py --dynamic-iterative "..."with round limits / auto-controller flags
Defaults in the UI favor iterative + auto where configured.
Recent behavior updates:
- Markdown answers render client-side with sanitized HTML (using bundled browser ESM vendor assets).
- File uploads are supported in chat; server writes files to
<tool>/_web_uploads/<uuid>/and passes--dynamic-attachmentstomain.py. - Upload safety limits are enforced server-side (per-file, total bytes, max file count).
- Thumbs up/down ratings emit
(agentic) run_rating_metaon stderr withattachment_fingerprint(combined MCP + skill ids; legacymcp_fingerprintalias). The learning loop consumes these on the next planner run.
Setup
cd agentic-orchestration-web
npm install
npm start
Python: Install agentic-orchestration-tool deps in a venv; the server prefers agentic-orchestration-tool/.venv/Scripts/python.exe (Windows) when AGENTIC_PYTHON is unset.
URLs
- Default:
http://127.0.0.1:3847/ - LAN: set
AGENTIC_WEB_HOST=0.0.0.0inagentic-orchestration-web/.env
Health endpoints and metadata:
GET /api/pingreturns instance/pid metadata (useful to verify restarts and active process).GET /api/agent-providersreturns provider catalog metadata used by the UI selector.
Scripts
| Script | Purpose |
|---|---|
start-web.ps1 / start-web.sh |
Foreground npm start with auto-restart |
start-web-bg.ps1 / start-web-bg.sh |
Detached server |
stop-web-bg.ps1 / stop-web-bg.sh |
Stop detached server |
Host/port for background starters follow the same .env as foreground.
Security
The server runs local Python with user-supplied text. Do not expose it to the internet without authentication and hardening.
Related
agentic-orchestration-web/README.md- Configuration —
AGENTIC_*web variables - Dynamic planning
- Agent skills — procedural playbooks attached to runs
- Sessions learning and knowledge base — ratings