Crate merlion_core 

merlion-core

Shared types for the Merlion Agent: chat messages, tool calls, the tool trait, and the agent loop that drives an LLM ↔ tool conversation.

This crate is intentionally provider-agnostic. merlion-llm plugs a concrete LLM behind LlmClient, and merlion-tools plugs concrete tools behind Tool.

Re-exports

pub use agent::Agent;

pub use agent::AgentEvent;

pub use agent::AgentOptions;

pub use approval::AllowAllApprover;

pub use approval::ApprovalDecision;

pub use approval::DenyAllApprover;

pub use approval::ToolApprover;

pub use curator::Curator;

pub use error::Error;

pub use error::Result;

pub use llm::LlmClient;

pub use llm::LlmRequest;

pub use llm::LlmResponse;

pub use llm::LlmStreamEvent;

pub use llm::Usage;

pub use message::Message;

pub use message::Role;

pub use message::ToolCall;

pub use message::ToolResult;

pub use tool::Tool;

pub use tool::ToolRegistry;

pub use tool::ToolSchema;

Modules

agent

approval

Tool approval — a gate the agent runs through before dispatching a tool call. Mirrors hermes’s approval.py/set_approval_callback pattern but shaped around an async trait so non-CLI surfaces (a future messaging gateway, an HTTP API) can plug in their own approver.

curator

The Curator — a small turn counter that decides when the agent should pause to extract memories from the recent conversation.

error

llm

message

tool