Request: public API to send messages to the base Cuppy agent chat (SSE + tool events)

[78design]

Feature request: expose a public API to send messages to the base Cuppy agent chat

Background

We built a Teable-powered web app (B2B lead/sourcing directory) and want to embed the real Cuppy agent (the one users chat with inside the Teable base UI — with its DB querying, web access and skills) directly in our app, instead of reimplementing a simplified agent on the app side.

Current state (verified)

These /base/{baseId}/chat/* endpoints work:

• POST /base/{baseId}/chat/create → creates a session (201, returns chatId)
• GET /base/{baseId}/chat/{chatId}/messages → reads messages
• GET /base/{baseId}/chat/history → chat history

Missing / blocked:

1. There is no endpoint to send a message. Tried POST /base/{baseId}/chat/{chatId}/messages, /generation, /stream, /completions, /ask, /message, /messages/stream, etc. — all return route-level 404 (Express Cannot POST ...), i.e. the route is not registered on the server.
2. POST /base/{baseId}/chat/{chatId}/warmup and GET /base/{baseId}/chat/{chatId}/generation/status return "You are not allowed to edit/access this chat" for the app token.
3. The feature is closed-source (not present in this repo's features/chat — that controller is only the chart-completions endpoint), and the public API catalog (587 routes) has no message-send route.
4. No embeddable Cuppy chat UI exists either (app.teable.cn sends x-frame-options: deny; Claw bots only support Feishu IM with no web channel).

Proposal

Expose a documented, streamable endpoint for third-party apps, consistent with the existing /base/{baseId}/chat/* style:

POST /base/{baseId}/chat/{chatId}/messages
Authorization: Bearer <app token or user session token>
{ "content": "user message text" }
→ SSE stream (text/event-stream):
   - content_block_delta (text_delta) for incremental text
   - tool-use / progress events so clients can render the agent's task timeline

Would appreciate clarity/availability on:

1. A message-send endpoint with streaming, keeping conversation context across turns (same chatId).
2. Tool/skill transparency: expose the agent's tool execution (DB queries, web fetch, installed skills) as SSE events.
3. Permission model: allow both app tokens and logged-in user session tokens to send (and open warmup/generation/status to the chat creator).
4. Fallback if a full agent endpoint is not feasible soon: an endpoint like POST /api/{baseId}/ai/generate-stream but with base context + tools.

Why it matters

• Lets third-party Teable apps reuse the real Cuppy (skills + tools) instead of reimplementing agents per app.
• Matches the in-UI Cuppy experience (thinking process, task timeline, skill calls).
• Happy to participate in testing once available.

We can reproduce the missing-route behavior instantly on request for debugging.