What Is Metateam

You run metateam in your project directory. A terminal dashboard launches. You tell the crew leader what you need — "fix the checkout bug," "write tests for auth," "refactor the API layer" — and a team of AI agents coordinates the work while you watch, steer, and review.

Every session is captured automatically. Knowledge accumulates across sessions. The next time any agent opens this project, it starts warm — with context from past work and your curated project conventions already loaded.

You (human)
  |
  |-- Terminal Dashboard (metateam) --> Crew of AI Agents --> CLI commands, code, tests
  |
  |-- Web Interface (metateam.ai)  --> Session history, KB management, API keys

The Problem It Solves

Every new session with an AI coding agent starts with amnesia. Context is rebuilt every morning. Shared understanding — the most expensive thing in development — evaporates.

Metateam fixes this with a persistent feedback loop: sessions are captured, knowledge is curated, and context is injected automatically into every future session.

What You Do vs. What Agents Do

You (the human) Agents (in their terminals)
Run metateam to launch the dashboard Work in individual terminal sessions
Tell the crew leader what you want Execute tasks, write code, run tests
Switch tabs to observe agent progress Coordinate with each other via messages
Steer agents directly when needed Read/write the knowledge base
Review results in mail-board reports Report back when work is done

Most CLI commands (crew message, kb set, memory recall, etc.) are designed for agents running inside their terminal sessions. You interact through the dashboard.

Core Features

  • Terminal dashboard with tabbed agent views, built-in messaging, and one-key crew controls.
  • Automatic session capture — every agent session is saved and searchable.
  • Warm start context injection — agents begin each session with relevant history and KB entries already loaded.
  • Hierarchical Knowledge Base for project conventions, architecture decisions, and gotchas.
  • Semantic memory — facts extracted from sessions, searchable by meaning.
  • Multi-agent crews — summon specialists, coordinate through a team leader, review through mail-board reports.
  • DOT-defined pipelines (Attractor) for repeatable multi-stage workflows.
  • Local code search using FTS5/BM25 and regex.
  • Web interface at metateam.ai for browsing sessions, managing the KB, and account settings.

Next Steps