Hall Chat MVP - Multi-Agent Collaboration for ClawUI

Status: Draft | Priority: High | Target: v1.0 | Date: 2026-03-31

Overview

Hall Chat enables multiple AI agents to collaborate on tasks in a shared workspace — similar to a team chat room, but with structured task cards, execution queues, and handoff protocols. Reference Implementation: OpenClaw Control Center’s Collaboration Hall (/src/runtime/collaboration-hall-*)

Problem Statement

Currently in ClawUI:
  • Each agent works in isolation
  • Agents cannot delegate tasks to each other
  • No shared context or task coordination
  • Users manually copy information between agents
We need: A shared “hall” where agents can discuss, delegate, execute, and hand off work in a structured way.

Core Concepts

Hall

A virtual collaboration space for a project or task. Contains:
  • Participants (agents + operator)
  • Task Cards
  • Message History

Task Card

A unit of work tracked in the hall:

Participant

An agent or human in the hall:

Message

A communication in the hall:

MVP Features (v1.0)

P0 - Must Have

P1 - Should Have

P2 - Nice to Have


User Flows

Flow 1: Discussion → Execution

Flow 2: Handoff Between Agents


Technical Architecture

Components

Data Model

Hall
HallMessage (stored in Hall)
TaskCard (stored in Hall)

API Endpoints

NATS Subjects


Implementation Phases

Phase 1: Foundation (MVP)

  • Hall data model
  • Hall Store (in-memory or Redis)
  • Basic REST API
  • SSE streaming
  • Create/list/delete halls
  • Add participants
  • Post/list messages

Phase 2: Task Coordination

  • Task Card CRUD
  • Stage transitions
  • Execution queue
  • @mention routing
  • Basic handoff

Phase 3: Agent Integration

  • Runtime dispatch to OpenClaw agents
  • Agent session linking
  • Streaming responses from agents
  • Structured handoff protocol

Phase 4: Polish

  • Hall summaries
  • Artifact sharing
  • Custom hall rules
  • UI components refinement

UI Mockups (Text)

Hall List View

Hall Detail View


Open Questions

  1. Persistence — Redis only, or PostgreSQL?
  2. Agent Discovery — How do we know which agents exist?
  3. Authorization — Who can create halls? Add participants?
  4. Scaling — Multiple halls vs single hall per project?
  5. History — How long to keep messages?

Reference Files

These are the source files from OpenClaw Control Center that implement Hall Chat:
Key reading order:
  1. hall-runtime-dispatch.ts — How agents get dispatched and return responses
  2. collaboration-hall-orchestrator.ts — Main flow and state machine
  3. hall-handoff.ts — Structured handoff format
  4. hall-speaker-policy.ts — Discussion queue management

Last updated: 2026-03-31