Agentfield

AgentField integrates Pydantic or Zod schemas with LLM calls to return typed, validated outputs, as shown in the code example where `app.ai(schema=Decision)` ensures structured decision-making.

It provides a control plane for routing calls between agents with full tracing and discovery, enabling complex multi-agent workflows like those in the autonomous engineering team example.

Every execution generates a tamper-proof, offline-verifiable credential using W3C DIDs, ensuring non-repudiation and governance, which is critical for production systems.

Supports fire-and-forget execution with webhooks and allows pausing for human approval via `app.pause()`, making it suitable for long-running, crash-safe tasks like claims processing.