See It in Action
A master agent (Claude) decomposes the goal, sub-agents (Codex, opencode) implement in parallel, and the monitor panel on the right keeps everything in view — a real session.
Click the image to enlarge
Agent Orchestrator
Command a team of AI coding agents in parallel, from one window
Every task runs on its own isolated git worktree and branch. Delegate a big goal and a master agent breaks it into sub-tasks that run in parallel. From scheduled runs to verification and merge — an orchestration tool for operating a team of agents.
A master agent (Claude) decomposes the goal, sub-agents (Codex, opencode) implement in parallel, and the monitor panel on the right keeps everything in view — a real session.
Click the image to enlarge
Not a tool for driving one agent — a tool for operating a team of agents.
Every task gets its own git worktree and branch, created automatically. Whatever an agent changes, your main branch stays safe — review the diff and merge with one click. Finished worktrees clean themselves up.
Give a master agent a big goal; it decomposes it into sub-tasks and runs them in parallel. Route each kind of work — coding, research, tests, docs — to a different vendor.
Claude tasks run in the official Claude Code TUI you can see and type into, while the task pane lays out progress, tool runs, and results. The extension never handles your Claude login.
Schedule a task and the tree shows the time remaining; missed schedules start when VS Code reopens. Pair it with the desktop app and schedules run even when VS Code is closed.
A read-only verification session cross-checks instructions, diff, and report for a pass/fail verdict, and the gate runs your build and tests after each sub-task, sending new failures back before merge. (opt-in)
Switch a task to another agent or profile and continue on the same worktree. A sub-task stopped by a usage limit or auth error moves on to a fallback profile automatically.
Run and monitor milestones and tasks across multiple git repositories from one multi-root window — with different vendor profiles per repository, at the same time.
Pair another computer or your phone by code or QR and check logs, send instructions, approve, and merge remotely. Phones connect through the browser (app.100doo.com); execution stays on the original machine.
Discussion-only sessions that never touch code. Two or three agents debate in rounds, and the conclusion drops straight into a new task's instructions.
Bundle role prompts and extra rules into packs, and save run defaults, models, and permissions as profiles you can switch per repository.
Track per-vendor cost, success rate, and trends, and line models up on cost, duration, error rate, and merge rate. Run metrics never leave your machine.
Execution, git changes, and scheduling only work in trusted workspaces. API keys live in VS Code SecretStorage, and branches that touch sensitive files such as hooks or settings are never auto-merged.
Labs features are switched on from the extension's Labs panel (Agent: Open Labs).
From milestone creation to merge — five steps are all it takes.
Create a milestone from the 🤖 icon in the activity bar. If the folder isn't a git repository yet, onboarding walks you through git init.
Run a single task from one instruction, or delegate a big goal to a master agent for parallel decomposition.
Watch progress in the task pane, keep instructing, and approve. Delegated runs show master and subs in one view.
The verification agent and the subtask gate (build and tests) check results before merge. (opt-in)
Review the diff and merge the task branch into the base branch. Delegated tasks support batch merge too.
All eight built-in vendors work in every role — standalone, master, and sub — and any Anthropic-compatible server lets you add local models of your own. Mix and match vendors per role.
| Vendor | Runs via | Roles |
|---|---|---|
| Claude | Official Claude Code TUI (default) · Agent SDK · CLI | StandaloneMasterSub |
| Codex | Codex CLI | StandaloneMasterSub |
| DeepSeek | Anthropic-compatible endpoint (reuses claude CLI) · API key | StandaloneMasterSub |
| GLM | Anthropic-compatible endpoint (reuses claude CLI) · API key | StandaloneMasterSub |
| Solar | Upstage Anthropic-compatible endpoint (reuses claude CLI) · API key | StandaloneMasterSub |
| Grok | grok CLI · ACP (default) / streaming-json | StandaloneMasterSub |
| Antigravity | agy CLI (stream-json) · OAuth — no API key needed | StandaloneMasterSub |
| opencode | opencode CLI · ACP | StandaloneMasterSub |
| Custom (local models) | Any Anthropic Messages API server (reuses claude CLI) · LM Studio, Ollama, OpenRouter presets | StandaloneMasterSub |
Register API keys and custom vendors in the Agent: Manage Vendors panel. A per-vendor concurrency limit protects low-spec local servers.
Each works fully on its own; together they give you 24/7 oversight.
The main stage for interactive runs, approvals, diffs, and merges
Always-on menu-bar oversight + runs without VS Code
The two programs share nothing but the filesystem inside your repository — no IPC, no server — and a file lock plus a schedule ledger make each schedule fire exactly once. Install either one and it works completely on its own.
Install the VS Code extension and the macOS desktop app.
Search for 100doo in the VS Code Extensions tab (⇧⌘X), or from the terminal:
View on VS Code Marketplace → View on Open VSX →
After installing, run Agent: Doctor from the Command Palette (⇧⌘P) to check CLI installs and logins in one go.
Download the DMG for your Mac's chip, open it, and drag 100doo into the Applications folder. When it starts, the monitor panel opens and a 100doo icon appears in the menu bar at the top of the screen — not in the Dock. Later versions are downloaded by the app and installed when you quit it.
Free · Apache-2.0 license · agent usage is billed by each vendor's own subscription or API pricing · source code planned to open after v1.0
Send questions and ideas through the Marketplace Q&A, and desktop app bugs through GitHub Issues.