Crate daimon_core 

daimon-core

Core traits and types for the Daimon AI agent framework. This crate is the plugin interface — implement Model to add a new LLM provider.

Provider crates depend on daimon-core for the shared types and trait. The main daimon crate re-exports everything from here, so end users typically never need to depend on daimon-core directly.

Implementing a Provider

use daimon_core::{Model, ChatRequest, ChatResponse, Result, ResponseStream};

pub struct MyProvider { /* ... */ }

impl Model for MyProvider {
    async fn generate(&self, request: &ChatRequest) -> Result<ChatResponse> {
        // call your LLM API
        todo!()
    }

    async fn generate_stream(&self, request: &ChatRequest) -> Result<ResponseStream> {
        todo!()
    }
}

Re-exports

pub use distributed::AgentTask;

pub use distributed::ErasedTaskBroker;

pub use distributed::TaskBroker;

pub use distributed::TaskResult;

pub use distributed::TaskStatus;

pub use vector_store::ErasedVectorStore;

pub use vector_store::SharedVectorStore;

pub use vector_store::VectorStore;

Modules

distributed

Core types and trait for distributed agent execution.

stream_util

Shared, dependency-light helpers for chunked HTTP streaming and retry backoff, used by every reqwest-based provider.

vector_store

Trait-based plugin system for vector stores.

Structs

ChatRequest

A request to send to a model.

ChatResponse

A complete response from a model.

Document

A retrieved document fragment with optional metadata and relevance score.

Message

A single message in a conversation.

ScoredDocument

A document paired with a similarity score from a vector query.

ToolCall

A tool invocation requested by a model, containing the tool name and arguments.

ToolOutput

The result of executing a tool.

ToolRetryPolicy

Policy controlling when and how tool execution is retried on failure.

ToolSpec

Describes a tool that can be passed to a model.

Usage

Token usage statistics for a single model request.

Enums

BackoffStrategy

Strategy for computing delay between retries.

DaimonError

The central error type for all Daimon operations.

Role

The role of a participant in a conversation.

StopReason

Why the model stopped generating.

StreamEvent

An event emitted during a streaming model response.

ToolChoice

Controls which tools the model is allowed to use.

Traits

EmbeddingModel

Trait for models that produce vector embeddings from text.

ErasedEmbeddingModel

Object-safe wrapper for EmbeddingModel.

ErasedMemory

Object-safe wrapper for the Memory trait, enabling dynamic dispatch via Arc<dyn ErasedMemory>.

ErasedModel

Object-safe wrapper for Model, enabling dynamic dispatch via Arc<dyn ErasedModel>.

ErasedTool

Object-safe wrapper for the Tool trait, enabling dynamic dispatch via Arc<dyn ErasedTool>.

Memory

Trait for conversation memory backends. Stores and retrieves messages for agent context.

Model

Trait for LLM providers. Supports both synchronous and streaming generation.

Tool

Trait for tools the agent can invoke. Tools must have unique names and declare a JSON Schema for parameters.

Type Aliases

ResponseStream

A boxed, pinned stream of StreamEvent results.

Result

A type alias for Result<T, DaimonError>.

SharedEmbeddingModel

Shared ownership of an embedding model.

SharedMemory

Shared ownership of memory via Arc<dyn ErasedMemory>. Used by the agent.

SharedModel

Shared ownership of a model via Arc<dyn ErasedModel>.

SharedTool

Shared ownership of a tool via Arc<dyn ErasedTool>. Used by registry and agent.