Value proposition

One coding agent is a chat window. Five agents is an operations problem. ainb is built to solve that operations problem.

Problems solved

Challenge	Without ainb	With ainb
Branch clobbering	Agents share one working directory, corrupting uncommitted files and colliding on branches.	Every session gets an isolated git worktree directory and branch (agents/<hash>).
Session survival	Closing a terminal window or laptop sleep kills the running agent.	Agent runs in a background tmux session. Detach with Ctrl-b d, reattach with ainb attach <name>.
Human in the loop	You cycle through terminal tabs constantly checking if agents need input.	Central Fleet overlay (f) and Inbox (b) display all pending ASK and APPROVE requests in one place.
Code verification	You blindly commit or run external diff tools to see what the agent changed.	Warp-style hunk-by-hunk diff review (g) with collapsible files and word-level syntax highlighting.
Cost attribution	Monthly API invoices show a single lump sum across all repos.	burndown tracks token spend per project, per model, and per day from local JSONL logs.
Tool drift	Rules and skills copied across ~/.claude, ~/.codex, and ~/.copilot drift out of sync.	ainb-toolkit writes skills and agents once, deploying to 9 tools from a single source.

What it costs

• Zero subscriptions: Open source (MIT), runs completely on your local machine.
• Zero telemetry: Does not phone home. The only network calls are those made by your AI providers.
• Zero lock-in: Controls tools through standard PTYs, not proprietary wrapper SDKs.

Operational boundaries

• Worktree isolation isolates git state, not system services: Running agents still share local ports (e.g. 3000) and databases on your machine.
• Prerequisites: git (2.30+) and tmux (3.2+) are required.
• Windows: Supported through WSL2.

Next steps

• Overview: monorepo architecture and component breakdown
• Install: installation methods
• Quickstart: launch your first session in 60 seconds
• Concepts: workspaces, worktrees, and attention