Ultra-lightweightMulti-agent OrchestratorAnywhere you work
Write plan. Run plan. Workbench does the rest.
$pip install wbcli
terminal
$ wb run plan.md
Plan: API Authentication System
Tasks: 5 across 3 wave(s)
Concurrency: 4
—— Wave 1/3 (3 tasks) ——
Workbench
| Task | Status | Time | Pipeline |
|---|---|---|---|
| User model | done | 3m42s | impl:ok → test:pass → review:pass |
| Auth tokens | done | 4m18s | impl:ok → test:fail → fix → test:pass → review:pass |
| Shared types | done | 2m05s | impl:ok → test:pass → review:pass |
Merging 3 branch(es) into scaffold...
✓ wb/user-model — Merged cleanly.
✓ wb/auth-tokens — Merged cleanly.
✓ wb/shared-types — Merged cleanly.
—— Wave 2/3 (1 task) ——
| Task | Status | Time | Pipeline |
|---|---|---|---|
| Auth middleware | running | 2m31s | impl:ok → test:pass → review... |
—— Wave 3/3 (1 task) ——
| Task | Status | Time | Pipeline |
|---|---|---|---|
| API endpoints | waiting | — | — |
How it works
Write a plan
Describe tasks in a markdown file with dependencies and file ownership
Run one command
wb run plan.md dispatches agents in parallel
Ship the result
Agents implement, test, review, and merge — you review the PR
Everything you need to orchestrate AI agents
Multi-Agent Support
Claude Code, Gemini CLI, Codex, or your own custom agent
Parallel Execution
Independent tasks run simultaneously in isolated git worktrees
Dependency Waves
Automatic task ordering respects dependencies between work
TDD Mode
Tests first, implementation second — built into the pipeline
Smart Profiles
Configure agent roles, directives, and assignments per project
Auto-Merge
Parallel branches merged automatically with conflict resolution
From plan to code in minutes
plan.md
# Add user authentication
## Task 1: User model
agent: claude
files: src/models/user.ts, src/db/schema.ts
Create the User model with email, password hash,
and session fields. Add migration.
## Task 2: Auth endpoints
agent: claude
depends: task-1
files: src/routes/auth.ts
Implement login, register, and logout
endpoints using the User model.
terminal
$ wb run plan.md
workbench v0.9.2
Wave 1 — 2 tasks in parallel
✓ user-model (claude) done 42s
✓ product-model (gemini) done 38s
Wave 2 — 1 task
✓ auth-endpoints (claude) done 1m 15s
✓ All tasks complete
3 branches merged into main