Hacker News new | ask | show | jobs
Show HN: Metateam: run many Claude/Codex/Gemini CLI instances in one terminal UI (metateam.ai)
3 points by falsename 107 days ago 

  Single Rust binary. Manages multiple AI coding agent sessions in tmux — Claude Code, Codex CLI, Gemini CLI side by side. Dashboard gives you tabs (F1–F11) to see each agent's terminal live.                                                                                           
                                                                                                                                              
  The agents get personas, persistent memory across sessions, and can message each other directly. Cross-machine P2P works over TLS 1.3 so crews on different machines collaborate.
  My favorite part: a dedicated archivist agent that never writes code, just indexes the entire repo on every push so other agents stop burning context grepping for files but ask him questions instead.
Free account, created during first run.
2 comments
nachocoll 103 days ago
The "archivist agent that never writes code, just indexes the entire repo on every push" is the most interesting part of this system. It solves the context-grepping problem — agents burning tokens trying to understand codebase structure before they can do useful work — by making structural knowledge a queryable service rather than a repeated scanning task.

The Knowledge Base with versioned hierarchical key-value store for architectural decisions is essentially a structured CLAUDE.md on steroids. When agents can read and write to shared architectural context, the team's decisions become persistent inputs to generation rather than things that have to be re-established in every session.

The Agile Vibe Coding Manifesto's principle that "context is explicit and versioned" directly addresses the problem you're solving. The manifesto's argument is that architecture decisions, crew instructions, and codebase maps are first-class artifacts of the development process — not documentation as afterthought, but the scaffolding that makes coherent multi-agent generation possible.

The cross-machine P2P coordination is fascinating — what happens when two agents on different machines make conflicting architectural decisions and try to write them back: https://agilevibecoding.org

link
falsename 107 days ago
Stack:

  - CLI + dashboard: Rust, single binary. TUI built with ratatui, agents run in tmux sessions. Full ANSI rendering — you see exactly what each agent shows.
  - API: ASP.NET Core minimal API on .NET 10, SQLite for everything (sessions, KB, users, settings).
  - Memory: sqlite-vec for vector search + FTS5 for full-text. Embeddings via Ollama (local). Facts auto-extracted from session transcripts by LLM, recalled on session start via semantic similarity.
  - P2P: Direct peer-to-peer over TLS 1.3 between machines, with relay fallback for NAT. Agents on different machines message each other and transfer files without going through a central server.
  - Session hooks: SessionStart/SessionEnd hooks inject memory context and capture transcripts automatically. Supports Claude Code, Codex CLI, and Gemini CLI — each has its own transcript parser (JSONL, markdown, Gemini format).
  - Knowledge Base: Hierarchical key-value store (project/compartment/entry), versioned, API-backed. Agents read/write it. Crew instructions, architecture decisions, codebase maps all live here.
No containers, no background services beyond tmux.

Check the blog section about the recently added Archivist for a short video showcase.

Happy to answer questions here.

link