feat: Chat Sessions — persistent conversational chat with agents

[OneStepAt4time]

Chat Sessions — Persistent Conversational Chat with Agents

Why

Lower friction interaction with agents. Currently users must create a session, send a message, poll for transcript. Chat sessions make it feel like iMessage — open a conversation, type, get a response. No setup, no babysitting.

Concept

• A Chat Session is a persistent conversation between a user and an agent
• Messages go back and forth naturally (user → agent → user)
• Each chat message creates an agent task in the background
• Chat sessions persist across messages, like a messaging thread

DB Schema

CREATE TABLE chat_session (
  id TEXT PRIMARY KEY,
  workspace_id TEXT NOT NULL,
  agent_key TEXT NOT NULL,
  creator_id TEXT NOT NULL,
  title TEXT NOT NULL DEFAULT '',
  status TEXT NOT NULL DEFAULT 'active' CHECK(status IN ('active', 'archived')),
  created_at TEXT NOT NULL DEFAULT (datetime('now')),
  updated_at TEXT NOT NULL DEFAULT (datetime('now'))
);

CREATE TABLE chat_message (
  id TEXT PRIMARY KEY,
  chat_session_id TEXT NOT NULL REFERENCES chat_session(id) ON DELETE CASCADE,
  role TEXT NOT NULL CHECK(role IN ('user', 'assistant')),
  content TEXT NOT NULL,
  task_id TEXT,  -- links to agent task if applicable
  created_at TEXT NOT NULL DEFAULT (datetime('now'))
);

API Endpoints

• POST /v1/chat — create chat session (pick agent)
• GET /v1/chat — list user chat sessions
• GET /v1/chat/:id — get session with messages
• POST /v1/chat/:id/messages — send message (triggers agent response)
• DELETE /v1/chat/:id — archive session
• PATCH /v1/chat/:id — update title

Real-time

• SSE events for new chat messages
• Typing indicator when agent is processing
• Session read/unread state

Frontend

• Chat page: session list sidebar + message thread
• Message bubbles: user on right, agent on left
• Input bar at bottom with send button
• New chat button: pick agent, start conversation

Acceptance Criteria

• CRUD API for chat sessions and messages
• Send message triggers agent task and returns response
• SSE events for new messages
• Dashboard chat page with session list and message thread
• Tests for API endpoints
• npm run gate passes

Reference

Competitive analysis: references/multica-competitive-analysis.md §2.4

[OneStepAt4time]

🚧 Blocked by: #3970 (Task Queue) — foundational state machine required before this feature can be implemented.

[OneStepAt4time]

🚧 Blocked by: #3971 (Agent Profiles) — first-class agent entities required before this feature can reference them.