Autonomous development that never stalls.

An orchestration system that coordinates multiple AI agents to autonomously run from requirements to implementation, judgement, and retries.
Designed for failure recovery, prioritizing convergence to completion.

Start Orchestration
$ git clone git@github.com:Andyyyy64/openTiger.git
openTiger — orchestration
git clone git@github.com:Andyyyy64/openTiger.git
cd openTiger
pnpm install
Packages: +1280
Done in 12.4s
cp .env.example .env
pnpm run up
Planner: started
Dispatcher: started
Workers/Testers/Docser: ready
Judge: started
Cycle Manager: started
Dashboard: http://localhost:5173
_

Engineered for
Inevitability.

Traditional agents stop when they fail. openTiger is built to recover, retry, and converge.

Completion First

Completion rate in long-running operations is the priority. The system is designed around the assumption that failures will happen, ensuring eventual consistency through autonomous recovery loops.

FAILURE RETRY COMPLETION

Lease-Centric

Parallel control for long-running workloads using a lease mechanism to manage resource contention and agent lifecycles.

Worker A Worker B Worker C

Idempotent Judgement

Judgement criteria are machine-verifiable and idempotent, preventing endless loops of subjective reviews.

PASS

Non-Destructive Verification

Testing and verification occur in isolated environments, ensuring the project state is never mutated until success is confirmed.

if (verification.passed) {
merge(changes);
} else {
revert();
}

Autonomous Cycle

Planner Dispatcher Execution Judge Done Cycle Manager (Retry)

Make failure boring.

Start with the docs, explore the architecture, and ship a workflow that always converges.