Agentic patterns

Drive an agent by alternating reasoning and acting steps inside a structured loop that terminates when the model signals a final answer.

Register typed tools by name so an agent can dispatch LLM-selected function calls to the right handler without embedding dispatch logic in the agent itself.

Decompose a complex task by having a root agent spawn and coordinate specialized subagents, each responsible for one concern, then aggregate their results.

Give an agent a pluggable memory store — short-term context buffer, episodic session history, and optional long-term semantic store — through one interface so the storage backend can be swapped.

Ground LLM responses in authoritative documents by retrieving relevant chunks before generation so the model can cite real data instead of hallucinating.

Pause agent execution at defined checkpoints to request human approval or input, then resume with the human's decision injected back into the agent's context.

Build prompts from versioned, parameterized templates so prompt logic stays separate from agent orchestration code and can be tested, diffed, and swapped independently.

Enable agents to exchange typed messages through a shared message bus so they can collaborate without direct references to each other's implementations.