org.llm4s.effect.cats

cats-effect wrapper for Agent.

Lifts every Result[AgentState] return value into F[AgentState], raising LLMError as LLMException in the error channel. The underlying Agent.run and related methods are blocking — each call is dispatched to the blocking thread pool via Async[F].interruptible, so cancelling the fiber interrupts the running agent loop.

Intentionally a thin wrapper: run does not expose handoffs, and continueConversation does not expose contextWindowConfig; the underlying Agent defaults apply. Tracing, debug logging and the trace log path are still available through the context parameter. Use Agent directly (inside Async[F].interruptible) when you need handoffs or context-window pruning.

Tool calls are not a failure of the effect: Agent hands a call with invalid arguments back to the model as a structured error tool result and the run continues. The experimental schema-validated AgentTool contract and the graph runtime (org.llm4s.agent.graph) are not wrapped here.

cats-effect wrapper for LLMClient.

All blocking LLM calls are shifted to the blocking thread pool via Async[F].interruptible, keeping the compute pool free for fibers. Cancelling the fiber (or a timeout) interrupts the provider thread, which is how llm4s providers are cancelled: they keep the interrupt and return Left(CancelledError). LLMError values are surfaced as LLMException in the F error channel.

Bridges LLMError into the cats-effect error channel as a Throwable.