A DAG-based parallel workflow engine that routes tasks across Claude, Ollama, and TaskFlow AI — with real-time cost tracking, automatic failover, and autonomous worker mode. Built in Go for production workloads.
Tasks scheduled as a directed acyclic graph with automatic dependency resolution. Independent tasks run concurrently across workers, delivering 1.3-3x speedup on production workloads.
Intelligent priority routing across Claude API, Ollama, and TaskFlow AI. Auto-selects Haiku for simple tasks, Sonnet for medium, Opus for complex. Graceful fallback through the provider chain.
Every LLM call is metered with per-task cost attribution. Set cost ceilings to halt execution when budget is exceeded. Route to free local models automatically.
Continuous daemon that polls TaskFlow for pending tasks, claims them, executes with multi-LLM routing, and reports results. Supports horizontal scaling with multiple workers.
Built-in research agent with Qdrant and Chroma vector store clients for retrieval-augmented generation. Scout engine explores solution spaces before committing to execution.
Structured span tracing for every workflow. LLM call logging, metrics collection, and WebSocket hub for real-time dashboard streaming.
Builds a DAG from your objective, identifies independent branches, and runs them concurrently. The scheduler handles dependency ordering, resource limits, and result aggregation automatically.
1.34x speedup measured on 10-task production workloads with 3 parallel workers.
Chains Claude, Ollama, and TaskFlow AI with priority-based routing. If Claude hits a rate limit, execution continues on Ollama. Your workflow never stalls.
Auto-selects the cheapest model capable of each task. Set a cost ceiling and the engine halts or routes remaining tasks to free local models.
Ollama on local hardware: $0.00 per task.
Runs as an autonomous daemon. Workers poll TaskFlow, claim tasks atomically, execute with full LLM routing, and push results back. Scale by launching more workers.
95.2% success rate across 42 autonomous tasks in benchmarks.
A DAG models task dependencies as a graph. Camshaft AI decomposes objectives into tasks, identifies independent branches, and runs them concurrently — eliminating idle time and delivering 25-300% speedup.
Yes. Configure Ollama as the primary provider and everything runs on your hardware at zero cost. The engine works identically with local or cloud models.
Every LLM call logs provider, model, token counts, and dollar cost. Set a cost ceiling per execution — the engine halts or reroutes to free models before exceeding budget.
Any task an LLM can decompose: code generation, research, document drafting, data transformation, API workflows, test generation, and multi-step reasoning chains.
Run complex AI workflows with parallel execution, multi-LLM routing, and real-time cost tracking.