ai-dev-os

AI Dev OS — developer-facing CLI and observable orchestration engine (ENGINE-006).

Structure

src/
├── core/           engine.py, context.py, bootstrap.py
├── agents/         base.py, openai_agent.py, claude_agent.py, gemini_agent.py, factory.py
├── workflow/       workflow_engine.py
├── tasks/          task_system.py
├── router/         agent_router.py
├── execution/      execution_layer.py, circuit_breaker.py
├── evaluation/     evaluation_engine.py
├── memory/         memory_store.py
├── observability/  logger.py, tracer.py, replay.py, metrics.py, record.py
├── cli/            main.py
└── plugins/        plugin_loader.py

Setup

pip install -e .
cp .env.example .env
# Edit .env with your API keys

Run

pip install -e .
cp .env.example .env
# Edit .env with your API keys

# CLI (recommended)
ai run --goal "Build a Subnet15 miner optimizer"
ai trace
ai replay
ai metrics
ai debug

# Or direct module
python -m src.core.bootstrap

Core loop

Goal → Workflow → Tasks → Agent Router → Circuit Breaker → Execution → Evaluation → Memory → Observability

Observability features

• Structured JSON logging for every system event
• Full execution trace per run (agent selection, execution, evaluation)
• ExecutionRecord data model for replay and debugging
• ReplayEngine to re-inspect decision flow
• Metrics collector (task_latency, agent_success_rate, evaluation_score, failure_rate, retry_count)

Reliability features

• Timeout control per agent execution
• Retry with fallback agent chain (claude → openai → gemini)
• Circuit breaker blocks agents after repeated failures
• Parallel async task execution
• Failure classification (timeout, model_error, network_failure, etc.)

Agent routing

Task	Model
analyze	Claude
plan	OpenAI
execute	Gemini