Free beta · runs on your machine

Your AI software team. One Team Lead.

Tell BeeBox what you want to build. Your Team Lead assembles the right team and gets it done.

Download BeeBox See how it works
Your project checkout stays on your machine Claude Code · Codex · Gemini · Qwen · Ollama Control it from your phone
Illustrative animation: you ask for a change; the Team Lead organizes agents and tasks, which move through review and QA. If the board has an integration stage, it can merge completed work into the checked-out base branch.
Agents run onClaude CodeOpenAI CodexGemini CLIQwen CodeOllama
How it works

From one sentence to merged code

You describe the outcome. The Team Lead can turn it into tasks, route them through your configured stages and report progress. If you configure an integration stage, it can merge completed work into the checked-out base branch.

  1. You

    Describe the outcome

    In plain words, in a chat — paste a screenshot if it helps.

    Add Face ID sign-in and rate-limit the login API.
  2. Team Lead

    Works out what it takes

    Inspects relevant project files and git history to investigate the request, then can create tasks with acceptance criteria and dependencies.

  3. Your team

    Assembled for the job

    Developers, a reviewer, QA, an architect or a designer — each role with its own model, access and number of agents working in parallel.

  4. Execution

    Work runs in parallel

    Every task gets its own git branch and worktree. Independent tasks run side by side; a task that depends on another waits for it.

  5. Done

    Checked and merged

    Review and QA stages check the work against the task. If configured, an integration stage rebases changes, runs its specified test command and fast-forwards into the checked-out base branch. The Team Lead reports back.

The deal

You manage the goal. BeeBox manages the agents.

No more starting agents one by one, copying context between them and checking who is stuck.

You

  • Say what you want — a feature, a bug, an audit.
  • Answer what only a person can — product decisions, access, money, anything irreversible.
  • Read the report — or check in from your phone.

BeeBox

  • Sets up the team — roles, models, access and how many agents work in parallel.
  • Plans the work — tasks with acceptance criteria, dependencies, priorities.
  • Runs it — schedules agents, uses separate worktrees for agent tasks and hands work from role to role.
  • Unsticks it — failed runs, loops and agents’ questions go to the Team Lead first.
  • Integrates it — review, QA and, if configured, a fast-forward into the checked-out base branch.
Operations View

See the team work. Step in when your input is needed.

The app’s live control room, rebuilt here with its real sections and labels: who is working on what right now, what waits for your decision, and what comes next.

PM / AnalystAnalysis · 1idle
DeveloperDevelopment · 3● T-12, T-13 11:12
Code ReviewerReview · 1◷ 1 queued
QA EngineerQA · 0idle
Merge into mainMerge into main · 1🔀 BeeBox merges
DoneDone · 14✓ 14

Now 2

T-12Face ID sign-in● RunningDevelopment
  1. ✓PM / Analyst
  2. →
  3. ●Developer
  4. →
  5. →
  6. →
🤖 Developer · Sonnet 5.5 · since 14:023 files changed·tests ✓·11 min
Now✏️ edits auth/biometric.dart
T-13Rate-limit the login API● RunningDevelopment
  1. ✓PM / Analyst
  2. →
  3. ●Developer
  4. →
  5. ○Code Reviewer↩
  6. →
  7. →
🤖 Developer · Sonnet 5.5 · since 14:092 files changed·4 min
↩ Code Reviewer sent it back · 1 issue(s) — return 429 with Retry-After instead of 500
Now▸ npm test -- login

Needs me 2

T-15Biometric sign-in on Android✋ Needs YouAnalysis
PM / Analyst asks: Should Android get fingerprint sign-in in this release, or iOS Face ID only for now?
Your answer…AnswerOpen
T-16Push notifications for new sign-ins✋ Needs YouDevelopment
Waiting for your approval to go on to Developer.
✓ ApproveOpen

Next 2

◷T-11Session refreshMerge into mainQueued · waiting to merge into main
⏸T-17Sign-in screen wiringDevelopmentWaiting · waits for T-12, T-13

Recent activity

14:09▶T-13 Developer started
14:08↩T-13 Code Reviewer ↩ Developer · 1 issue(s)
14:06❓T-15 PM / Analyst asks about Android
14:04→T-11 QA Engineer → Merge into main
14:02▶T-12 Developer started
13:58■T-10 QA Engineer finished
NowEvery running agent: the task, where it is on its way, the model, files changed, tests and what it is doing this second.
Needs YouOnly what really waits for a person: an agent’s question, an approval, a blocked task. Answer right in the card.
NextQueued and waiting work with the reason: a free agent, a dependency, the merge into main.
Capabilities

A team that fits the job, not the other way round

Everything below is in the app today.

The team forms around the goal

Ask for “two developers and a reviewer”, another model or a read-only role — the Team Lead adds roles from templates or your own prompt and applies the whole change in one step. It checks with you before adding an agent that writes code.

you › two developers and a reviewer on Opus lead › lead.configure_team Development: in parallel 1 → ×2 Reviewer: model → Opus ✓ applied

Any model, per role

Use the agent CLIs you already pay for. You can configure different models for roles and the Team Lead, subject to the providers you have set up: for example, a stronger model for review or a local model for triage.

Claude CodeOpenAI CodexGemini CLIQwen CodeOllama

Parallel work

Several agents on one role, independent tasks side by side, and parallel one-off jobs — an audit, research, the same job on several models to compare. Dependencies keep the order.

Review, QA, release

Tasks can carry acceptance criteria. Reviewers and QA can return work for unmet criteria, bugs, failing tests or security issues; configured loop limits cap review cycles. Add a release role with your own prompt.

Worktrees, then main

Agent tasks use their own branch and worktree. If configured, the integration stage rebases reviewed work, runs its specified test command and fast-forwards into the checked-out base branch; rebase conflicts are sent back with the conflicting file names.

Multi-repo projects

A project can include several directories — app, API, shared libraries. The Team Lead can inspect configured project directories; agents work in separate worktrees for repositories they change. The integration stage can merge those changes when configured.

It asks when your input is needed

Stuck tasks, failed runs and agents’ questions go to the Team Lead first. It can inspect task history and try to move work forward; it may ask you for product decisions, access, spending approval or other input.

A lead that knows your code

Ask “Where is this bug?” and the Team Lead can investigate project files, git history and configured read-only live checks, often pointing to relevant files and lines. Paste a screenshot with Ctrl+V. Prefer one agent? Solo mode can work on the request directly.

Local-first

Your project checkout stays on your machine

Agent processes run beside your checkout. With hosted providers, relevant prompts and code context are sent to that provider; local Ollama can keep model inference on your machine. BeeBox Cloud syncs organization workspace data and uploaded attachments, and relays phone connections.

  • Project files stay localBeeBox does not upload your repository, but agent prompts can send relevant code to your selected provider. Workspace data and uploaded attachments sync through BeeBox Cloud.
  • Bring your teamOrganizations share a workspace in real time; agents run on each teammate’s machine.
  • Coordinated task claimsA task claim lets one teammate’s machine run it at a time; projects are matched by git remote.
BeeBox architecture: agents run on each teammate’s machine next to the code; BeeBox Cloud syncs shared workspaces and relays the phone connection BeeBox Cloud sync · relay Your machine Team Lead · Sonnet Developer ×2 · Codex ● code on disk Teammate’s PC Reviewer · Opus QA · Ollama ● code on disk Phone Team Lead chat answers · approvals ● via desktop tunnel tasks · comments relay · QR
Mobile

Lead your team from your phone

Scan a QR code on your desktop and the phone becomes the remote: talk to the Team Lead with photos and files, answer agents, approve, follow live logs. The agents keep running on your machine; BeeBox Cloud relays the connection from any network.

  • Paired by a one-time QRNo passwords on the phone; the code lives five minutes.
  • Access to your machine onlyRevoke a phone from the desktop at any time.
  • Android app or the webA native Android app — or just your camera and browser.
Download

Give your Team Lead its first goal

You’ll need git and at least one supported agent provider configured: Claude Code, Codex, Gemini CLI, Qwen Code or a local Ollama service. Add your project folder and tell the Team Lead what to work on.

Choose Apple Silicon for an M-series Mac, or Intel for an Intel Mac. Android and the web app connect to BeeBox on your desktop.

FAQ

Frequently asked questions

What is BeeBox?

BeeBox is a desktop app for organizing AI coding agents around a project board. You describe a goal in chat; the Team Lead can break it into tasks with acceptance criteria and dependencies and route work through configured stages. Agents work in local project directories and git worktrees. If the board has an integration stage, BeeBox can rebase changes and fast-forward them into the checked-out base branch; a test command runs only if configured.

What does the Team Lead do?

It can inspect project files and git history to answer questions or investigate requested changes, and can use configured read-only live checks. For changes, it can create tasks and route them through the board; approval gates may require your input. You can ask it to adjust stages, models, access and parallel-agent limits. It may ask you for decisions or when blocked. In solo mode, it can work directly on the request.

How is BeeBox different from Copilot, Cursor or a chat assistant?

BeeBox adds a project board and Team Lead around agent CLIs. It can split a goal into tasks, run independent work in separate git worktrees, and route results through review and QA stages. An integration stage can merge work into the checked-out base branch when configured.

Do I have to configure the team myself?

A project can start from one of the board templates, including a full-cycle team with analysis, development, review, QA and integration stages. You can ask the Team Lead to adjust stages or models, and edit stages, agents, prompts, approval gates and limits in the board settings. Available templates and stages vary.

What do I need to get started?

A Windows, macOS or Linux machine with git and a supported agent provider set up: Claude Code, OpenAI Codex, Gemini CLI, Qwen Code, or a local Ollama service. Provider setup and billing are handled by that provider; BeeBox does not include provider usage.

Does my code go to the cloud?

The project checkout and git worktrees stay on your computer, but agent prompts can include relevant code and are sent to the provider you selected (unless you use a local model such as Ollama). For signed-in organizations, BeeBox Cloud syncs workspace data such as task text, comments and run summaries, and stores attachments you upload; it also relays the phone connection. Project files are not uploaded as a repository.

Can agents break things or push code?

Agent tasks normally use a separate branch and worktree. Agents are instructed not to push. If the board has an integration stage, BeeBox rebases changes and fast-forwards them into the checked-out base branch; it runs a test command only when one is configured, and pushes only when enabled. Approval gates and permission prompts can require your input.

Which AI models does BeeBox use?

BeeBox supports Claude Code, OpenAI Codex, Gemini CLI, Qwen Code and local Ollama models. The available models depend on the provider and your configuration; roles can be configured separately.

Can my whole team use it?

Organizations can share workspaces with real-time board sync. Agents run on each teammate's machine beside their own checkout. A task claim is coordinated so only one machine should run that task at a time; projects are matched across machines by git remote.

How much does it cost?

BeeBox is free during the beta. Agent provider subscriptions, API usage or local hosting are separate and depend on your setup.