Let your stajyer take over your job.

Orchestrate multiple AI coding agents without databases or heavy frameworks. You assign, they code, you review.

Get Started — it's free See how it works ↓

Works with:Claude CodeCodexCursor

AI Agents

Tasks Completed

115

Tests Passing

Total Time

“Continue” Commands

Sound familiar?

Running multiple AI agents sounds great until you hit these walls.

The Continue Tax

Terminal 1: ✓ Done. Waiting...
Terminal 2: ✓ Done. Waiting...
Terminal 3: ✓ Done. Waiting...

You: *switching tabs again*
     "continue" ↵

Your AI agents finish work and wait. You become the bottleneck.

File Wars

Error: File has been modified
       since read
Error: File has been modified
       since read
Error: File has been modified
       since read

Multiple agents editing the same file. Changes overwrite each other.

The Oops Moment

Agent: Moved dashboard, rewrote
  middleware, changed 5 files.
  Done! ✓

You: ...that was supposed to
     stay where it was.

No guardrails. An agent makes a sweeping change. You clean up the mess.

Stajyer fixes this.

A lightweight daemon that orchestrates your AI agents with three simple primitives.

Auto-Continue

Agents finish. Daemon dispatches. Zero human intervention. No more switching terminals to type “continue”.

stajyer up

[14:30] lead     ▶ Auditing codebase...
[14:32] lead     ✓ Created 5 tasks
[14:32] daemon   → Assigned 001 → frontend
[14:32] daemon   → Assigned 002 → backend
[14:35] frontend ✓ Completed: error-boundaries
[14:35] daemon   → Assigned 003 → frontend
[14:38] backend  ✓ Completed: api-tests
[14:38] daemon   → Assigned 004 → backend

Ownership Guard

Each agent has a territory. Protected files need lead approval. No more file conflicts.

.stajyer/config.yml

agents:
  frontend:
    owns: ["src/components/**"]
  backend:
    owns: ["src/app/api/**"]

protected:
  - path: "src/middleware.ts"
    requires: "lead"

Markdown State

State lives in markdown. Track progress in git. Read it in your editor. No database required.

.stajyer/tasks/

.stajyer/tasks/
├── 001-error-boundaries.md   ← git tracked
├── 002-api-tests.md          ← human readable
└── 003-skeleton-states.md    ← agent writable

Three steps. That's it.

Get from zero to orchestrated agents in under 5 minutes.

Init

terminal

$ npx stajyer init
# Creates .stajyer/ with config, task board, agent profiles

Hire

terminal

$ stajyer hire lead    # The coordinator
$ stajyer hire dev     # The workers (Claude Code, Codex, Cursor)

Let them work

terminal

$ stajyer up
$ stajyer task "Build the login page" --to frontend
$ stajyer task "Write API tests" --to backend
# Go grab coffee. ☕

“3 AI agents. 17 tasks. 115 tests. 2 hours.
Zero ‘continue’ commands.”

— The experiment that started Stajyer

Read the full experiment report →

How it compares

Most tools are single AI agents. Only a few orchestrate multiple agents simultaneously. Stajyer is the lightweight option that doesn't need a database.

Data sourced from public documentation, repos, and pricing pages. Last updated March 2026.

■ Multi-Agent Orchestrators— tools that coordinate multiple agents simultaneously

Feature	Stajyer	Paperclip	Parallel Code	Emdash (YC W26)	Composio	Stripe Minions
Open source	MIT	MIT	Yes	Yes	Yes	No (internal)
Setup time	5 min	15+ min	~10 min	~10 min	~10 min	N/A
Auto-continue	Yes	Yes (heartbeat)	No	No	No	Yes
File conflict prevention	Ownership guard	DB permissions	Git worktrees	Git worktrees	Git worktrees	Isolated devboxes
Task lifecycle	Yes (queue + dispatch)	Yes (issues + approvals)	No	No	Partial	Yes
Database required	No	Postgres (35+ tables)	No	No	No	Internal
State format	Markdown (git)	Postgres	None	None	None	Internal DB
Cross-tool agents	Claude + Codex + Cursor	Claude + Codex + Cursor	Claude + Codex + Gemini	Any provider	Any agent	Goose fork only
Available to public	Yes	Yes (self-hosted)	Yes	Yes	Yes	No

■ Single Agents with Partial Multi-Agent— can run multiple instances, but only of themselves

Agent	Multi-Agent	Cross-Tool	Open Source	Stajyer Adapter
Claude Code	Agent Teams (subagents within session)	No — Claude only	Source-available	✓ Supported
Codex CLI	Experimental (Agents SDK)	No — Codex only	Open source	✓ Supported
Cursor	Automations (cloud agents)	No — Cursor only	No	✓ Supported
GitHub Copilot	Mission Control (agent fleet)	No — Copilot only	No	Planned
Devin	Parallel sessions	No — Devin only	No	N/A (cloud)
OpenHands	Task delegation	No — OpenHands only	MIT (65k+ stars)	Planned

■ Single AI Coding Agents— powerful alone, better orchestrated with Stajyer

Agent	By	Type	Open Source	Stajyer Adapter
Aider	Paul Gauthier	CLI pair programmer	Apache 2.0	Planned
Cline	Community	VS Code extension	Apache 2.0 (5M+ devs)	Planned
Goose	Block	CLI agent	Apache 2.0 (29k+ stars)	Planned
SWE-agent	Princeton	Research agent	MIT	Planned
Continue.dev	Continue	IDE assistant	Apache 2.0	Planned
Sweep	Sweep AI	GitHub bot	MIT	Planned

The key difference: Worktree tools (Parallel Code, Emdash) isolate agents but don't manage them.

Stajyer adds what they're missing: auto-continue (no “continue” bottleneck), ownership guards (no file conflicts), and task lifecycle management (queue → dispatch → review). All without a database.

Free & Open Source

MIT licensed. No vendor lock-in. No limits.

RECOMMENDED

CLI

Stajyer CLI

$0 forever

Everything you need to orchestrate AI agents.

✓ Unlimited agents
✓ Auto-continue daemon
✓ Ownership guards
✓ Markdown state
✓ All adapters (Claude Code, Codex, Cursor)
✓ MIT licensed

coming soon

CLOUD

Stajyer Cloud

TBD

Web dashboard, team features, analytics.

○ Web dashboard
○ Team collaboration
○ Agent analytics
○ Templates marketplace

Stop being the bottleneck.

Your AI agents are waiting for you to type “continue”. Let Stajyer handle it.

Star on GitHub