Works with Claude Code, Codex, Gemini

Your AI agents,
coordinated.

Bee is your AI ops coordinator — always watching, so you don't have to.

Tell Bee what to build. She dispatches AI coding agents in isolated worktrees, monitors CI, forwards code reviews, and pings you on Telegram when your PR is ready to merge.

$ cargo install apiari
$ apiari

Who this is for

Built for developers who use AI coding agents and want to run more than one at a time.

Works with

Claude supported
Codex optional
Gemini optional
GitHub integration
Telegram optional
Sentry optional
Linear optional

How it works

Four steps. One command. Bee handles the rest.

Describe a task

Tell Bee what you need in plain English. She understands your codebase.

Bee dispatches a worker

Each worker gets its own git worktree. No conflicts, no babysitting.

Worker writes code + opens PR

The agent commits, pushes, and opens a pull request — all autonomously.

Bee notifies you when done

See everything in the TUI, or get a Telegram ping. Review and merge.

Your AI agents do the coding. Apiari does the ops.

Run multiple AI coding agents in parallel — each in its own isolated git worktree.

Signal hooks react to CI failures, reviews, and pushes — following playbooks automatically.

Choose Claude Code, Codex, or Gemini as your coordinator backend. Mix and match per workspace.

Get notified on Telegram when a PR is ready — step away and let it run.

Configure your workspace

Configure .apiari/ and Bee knows how to act, what to know, and what to do when things happen.

soul.md

Customize how the coordinator communicates. Set personality, tone, and communication style per workspace.

context.md

Load project knowledge into every session — tech stack, conventions, repo structure. Bee reads it before every task.

skills/*.md playbooks

Reusable runbooks loaded by signal hooks. Define how to handle CI failures, Copilot reviews, merge readiness, and more.

Signal hooks

Event-driven automation. When CI fails, a review comes in, or a PR is pushed, the coordinator follows a playbook automatically.

Authority model

autonomous gives full toolset access. observe is read-only. Fine-grained capabilities like merge_prs and dispatch_workers.

apiari ui TUI

Ratatui terminal dashboard with workers panel and coordinator chat. Connected to the daemon via Unix socket.

A day with apiari

Josh described a 4-feature sprint to Bee. She dispatched 4 workers in parallel — authentication refactor, API redesign, new test suite, and a UI overhaul. All 4 PRs landed with CI green and Copilot reviews addressed. Josh checked in the next morning.

The problem with running multiple agents

Agents conflict

Apiari isolates each agent in its own git worktree. No shared files, no merge conflicts, no corrupted state.

Too many tabs

One TUI to see everything. What's running, what's stuck, what opened a PR, what needs your attention.

Manual babysitting

Bee coordinates so you don't have to. She breaks down work, dispatches agents, and flags anything that needs a human.

Get started

Two commands. Bee will guide you through the rest. Works out of the box with Claude Code, Codex, Gemini, or any compatible agent.

Install + run

$ cargo install apiari
$ apiari
# That's it. Bee handles onboarding.

Just swarm (no coordinator)

Run agents in parallel without the coordinator. Manual mode.

$ cargo install apiari-swarm
$ swarm
# Press n to create a worker, type your task, go.
AI agent CLI

Required. Claude Code, Codex, Gemini, or compatible agent.

GitHub

Recommended. For PRs and repo integration.

macOS / Linux

Terminal-native. Works anywhere with a shell.

codex CLI

Optional. Alternative agent backend.

gemini CLI

Optional. Alternative agent backend.

The toolchain

Two binaries, one shared UI kit. Each does one thing well.

Under the hood

Architecture diagrams and live terminal demos for the technically curious.

Architecture

Two distinct flows — use swarm standalone for parallel agents, or add apiari for full orchestration.

Swarm: parallel agent execution

Apiari: full orchestration

Live terminal demos