Manage and Publish
Draft agent config in the UI, publish a version, and roll forward safely.
Manage and publish
Architecture: The Manage UI is a Clients layer interface to the backend's /api/manage routes — draft tools, policies, and models, then publish to the Persistence layer. See How CUGA works — Deployment.
CUGA is not only a demo. You can edit an agent in the browser, try it safely as a draft, then publish a version that production chat uses.
cuga start managerOpen http://localhost:7860. You get a manage UI where you can:
- Add or edit tools, MCP servers, and OpenAPI apps
- Pick the LLM and advanced model parameters
- Attach policies (including ToolGuard)
- Chat against the draft without changing what production users see
- Publish when the draft looks right — CUGA stores a versioned config so you can audit what shipped and roll forward
Draft chat is for you. Published chat is for everyone else. Publish only when you are happy with the draft.
Typical loop:
Start manager
cuga start managerEdit the draft
Change tools, MCP, model, or policies. Try a few tasks in draft chat.
Publish
Publish creates a new version. Production chat switches to that version.
When storage.mode = "prod", CUGA can keep existing published configs on startup instead of reseeding demo data. See preserve_configs_on_startup.
Related
- CLI presets — all
cuga startcommands - Demos — try sample apps before configuring your own
- Self-host on Kubernetes — deploy the published stack to a cluster
- Tool Approval policy — human-in-the-loop before sensitive tool calls