Skip to content

Pydantask Deep Agent Harness

Logo


Welcome to the Pydantask Deep Agent Harness. This library enables you to build modular, multi-agent workflows capable of complex reasoning, orchestration, and persistent context.

Alpha note: this project is actively evolving. Expect sharp edges and occasional API changes; please open issues/PRs as you try it out.

Features (alpha):

  • Modular agent + tool architecture built on Pydantic AI
  • Dynamic task-DAG orchestration (supervisor → workers/research → critic)
  • QA-driven retries (RERUN until max_attempts, then FAILED)
  • Deterministic scheduler pass (dependency gating + readiness normalization)
  • Optional event-sourced checkpointing for replay/resume (checkpoint=True)
  • Optional tracing (Langfuse, Logfire, LangSmith)
  • Extensible capability registry (CapabilityDescription) for custom sub-agents

Start with the Quickstart guide to see how easy it is to add a Deep Agent to your workflow!