🧠 Alinea – Multi-Agent Group Intelligence in Chat

Alinea is a multi-agent AI collaboration system built directly inside secure group chats.

Instead of a single assistant producing one answer, Alinea embeds multiple AI agents — each with a distinct reasoning style — inside an XMTP-powered Convos thread. The system structures breakout discussions, synthesizes perspectives, and produces a final coordinated recommendation in real time.

The project explores how human–AI collaboration changes when intelligence operates as a team rather than a single model.

Alinea multi-agent demo


🎥 Demo Video

Short live demo of the multi-agent discussion flow:


🔧 Tech Stack

  • TypeScript / Node.js
  • XMTP Agent SDK
  • Convos (XMTP group chat interface)
  • Kilo Gateway (OpenAI-compatible LLM API)
  • Multi-agent orchestration layer (custom)

⚙️ Core Features

  • Multi-agent collaboration inside group chat
  • Structured breakout discussions
  • Fixed-round deliberation mechanism (3–5 rounds)
  • Persona-based reasoning (Skeptic, Creative, Pragmatist, Coordinator)
  • Summary aggregation & final synthesis
  • Deterministic coordination script for demo stability
  • Persistent XMTP identity & encrypted messaging

🚀 Architecture Overview

High-Level Flow

User → Convos Group Chat
→ XMTP Network
→ Coordinator Agent
→ Breakout Threads
→ Summary Collection
→ Final Synthesis (Main Thread)


🧠 Agent Roles

🧩 Coordinator

  • Splits user prompt into structured breakout tasks
  • Orchestrates discussion rounds
  • Collects summaries
  • Produces final synthesized recommendation

⚠️ Skeptic

  • Identifies risks, trade-offs, weak assumptions

🎨 Creative

  • Generates bold, exploratory ideas

🛠 Pragmatist

  • Evaluates feasibility and implementation steps

🔄 Breakout Logic (Hackathon Scope)

To maintain stability and clarity:

  • Breakout groups are pre-configured XMTP threads
  • Each breakout runs a fixed number of exchanges
  • Convergence is time-based, not consensus-based
  • The coordinator follows a rigid script:

Split → Wait → Collect → Summarize → Synthesize

This ensures a demoable and deterministic multi-agent workflow.


🔐 Why XMTP?

XMTP provides:

  • End-to-end encrypted group messaging
  • Agent identities via wallet-based authentication
  • Scoped conversation boundaries

Agents operate as first-class participants inside the thread, not as an external overlay.


⚡ Key Design Principle

Traditional chatbots provide a single answer.

Alinea provides structured deliberation.

The system simulates collaborative reasoning in a shared environment, making the group context essential rather than decorative.


📂 Project Structure

src/

  • index.ts — XMTP agent bootstrap
  • coordinator.ts — Orchestration logic
  • breakout-agent.ts — Persona discussion engine
  • llm.ts — Kilo Gateway integration
  • personas/ — Agent identity prompts

🔗 GitHub Repository

Explore the source code and architecture:

👉 View on GitHub