# DagNats

> DAG-based workflow engine built on NATS for durable workflows and autonomous LLM coding pipelines.

## Key Concepts

- **Workflow**: A DAG of steps with dependencies, retries, timeouts, and concurrency limits
- **Step**: A unit of work (8 types: Normal, AgentLoop, SubWorkflow, Map, Sleep, WaitForEvent, Approval, Planner)
- **Worker**: A process that registers task handlers and consumes work from NATS
- **Run**: A single execution of a workflow with input, status tracking, and event history
- **Event**: An immutable record in the WORKFLOW_HISTORY stream driving DAG resolution

## Quick Start

1. `dagnats serve` — starts embedded NATS, engine, API, triggers
2. Define workflow as JSON with steps and dependencies
3. `dagnats workflow register workflow.json`
4. Write Go worker with `worker.NewWorker()` and `HandleTyped[I,O]()`
5. `dagnats run start <workflow> <input> --watch`

## Go SDK Packages

- `dag/` — Workflow definitions, builder API, DAG validation, pure Advance() state machine
- `worker/` — Task execution framework, TaskContext interface, typed handlers
- `protocol/` — Wire format types (Event, TaskPayload, TaskResolution)
- `observe/` — Provider-agnostic telemetry interfaces (Logger, Tracer, Metrics, ErrorReporter)
- `server/` — Embedded NATS server and lifecycle management
- `bridge/` — HTTP-to-NATS gateway for remote/non-Go workers
- `dagnatstest/` — One-call test setup with embedded NATS

## Documentation

Full docs at: https://github.com/danmestas/dagnats/tree/main/docs/site