Skip to main content
Back

The Vibe Company's Companion Manages Entire AI Agent Fleets Like Terraform

The-Vibe-Company/companion turns a folder of TOML files into a private fleet of specialized AI agents with memory, tools, and Tailscale-only access.

Pro

Repo, Full Access
  • Companion is a Go CLI that deploys fleets of specialized AI agents from declarative TOML workspaces.
  • Uses a Terraform-style validate, plan, apply loop with SQLite state and drift detection.
  • Manages Fly.io apps, Tailscale private networking, OpenRouter routing, and Granite memory vaults per agent.
  • Agents get identity via SOUL.md files, shared vaults for cross-agent memory, and a common Open WebUI.
  • Includes a live status dashboard, explicit destroy semantics, and a Fly-hosted control plane mode.
  • MIT-licensed at github.com/The-Vibe-Company/companion, sitting around 2.4k stars.

Managing a single coding agent is fine. Running five specialized agents with their own tools, memory, secrets, and private endpoints is where things get messy. Companion from The Vibe Company tackles that with a Terraform-style workflow: describe your agent fleet in TOML, preview a plan, then apply it to real infrastructure.

The project started as a web and mobile UI for Claude Code and Codex sessions, but the repo has since pivoted. The previous Companion web UI codebase is archived, and Companion is now a Go control plane focused on deploying fleets of agents rather than chatting with a single one.

From one agent to a fleet

Companion takes you from "I need several specialized agents" to a running fleet you can actually use. Each agent gets its own role, tools, memory, runtime, and private access. The bet is that tomorrow's agent stack looks less like one giant chatbot and more like a small team of narrow experts with distinct identities and permissions.

The tool manages these pieces per agent:

  • One Fly app, volume, secrets bundle, config render, and rollout per agent