Crate daimon 

Daimon

A Rust-native AI agent framework for building LLM-powered agents with tool use, memory, and streaming. Daimon implements the ReAct (Reason-Act-Observe) pattern: the agent calls a model, optionally invokes tools, observes results, and repeats until it produces a final response.

Quick Start

use daimon::prelude::*;

#[tokio::main]
async fn main() -> daimon::Result<()> {
    let agent = Agent::builder()
        .model(daimon::model::openai::OpenAi::new("gpt-4o"))
        .system_prompt("You are a helpful assistant.")
        .build()?;

    let response = agent.prompt("What is Rust?").await?;
    println!("{}", response.text());
    Ok(())
}

Feature Flags

Feature	Description
openai	OpenAI API provider (default, via daimon-provider-openai)
anthropic	Anthropic Claude API provider (default, via daimon-provider-anthropic)
macros	#[tool_fn] proc macro (default)
gemini	Google Gemini / Vertex AI provider (via daimon-provider-gemini)
azure	Azure OpenAI Service provider (via daimon-provider-azure)
bedrock	AWS Bedrock provider (via daimon-provider-bedrock)
ollama	Ollama local model provider (via daimon-provider-local)
llamacpp	llama.cpp (llama-server) provider (via daimon-provider-local)
llamars	llama-rs provider (via daimon-provider-local)
local	All local providers at once: Ollama, llama.cpp, llama-rs, generic OpenAI-compatible
sqlite	SQLite memory backend
redis	Redis memory backend + task broker
nats	NATS JetStream task broker
amqp	RabbitMQ (AMQP) task broker
sqs	AWS SQS task broker (via daimon-provider-bedrock)
pubsub	Google Cloud Pub/Sub task broker (via daimon-provider-gemini)
servicebus	Azure Service Bus task broker (via daimon-provider-azure)
a2a	Agent-to-Agent (A2A) protocol client (A2aClient)
mcp	Model Context Protocol client & server
otel	OpenTelemetry OTLP span export
qdrant	Qdrant vector store retriever
pgvector	pgvector-backed vector store (via daimon-plugin-pgvector)
opensearch	OpenSearch k-NN vector store (via daimon-plugin-opensearch)
grpc	gRPC transport for distributed execution
http-server	HTTP agent server (AgentServer)
eval	Agent evaluation/scoring harness
full	All providers + macros + MCP + SQLite + Redis + NATS + AMQP + OTel + HTTP server + Qdrant + pgvector + OpenSearch + gRPC + eval + SQS + Pub/Sub + Service Bus

The core framework compiles with no features; enable providers as needed.

Plugin Interface

The Model trait (from daimon_core) is the plugin interface. To create a new LLM provider, depend on daimon-core and implement Model. See the daimon-provider-* crates for examples.

Module Overview

• agent — Agent builder, ReAct loop, multi-agent patterns, resumable runs
• model — LLM provider trait and implementations
• tool — Tool trait, registry, and execution
• memory — Conversation memory, plus optional core/archival/episodic tiered memory subsystems
• stream — Streaming response types
• hooks — Lifecycle hooks for observability and control
• orchestration — Chain, graph, DAG, and workflow orchestration
• retriever — RAG retriever trait and tool adapter
• checkpoint — Checkpointing and state persistence
• a2a — Google Agent-to-Agent protocol support (feature = “a2a”)
• distributed — Distributed agent execution across processes
• mcp — Model Context Protocol client and server (stdio, HTTP)
• telemetry — OpenTelemetry OTLP export (feature = “otel”)
• server — HTTP agent server (feature = “http-server”)
• eval — evaluation harness for agent behavior (feature = “eval”)

Modules

a2a

Google Agent-to-Agent (A2A) protocol support.

agent

Agent construction and ReAct loop execution.

checkpoint

Checkpointing and state persistence for resumable agent runs.

cost

Token cost tracking and budget enforcement.

distributed

Distributed agent execution across multiple processes.

error

Error types for the Daimon agent framework.

eval

Agent evaluation and testing harness.

guardrails

Input and output guardrails for validating and transforming agent I/O.

hooks

Lifecycle hooks for observing and controlling agent execution.

mcp

Model Context Protocol (MCP) client and server.

memory

Conversation memory for persisting message history across agent turns, plus optional tiered memory subsystems that generalize the MemGPT/Letta-style memory model (core, archival, episodic).

middleware

Composable middleware pipeline for mutating requests, responses, and tool calls.

model

LLM provider abstraction and implementations.

orchestration

Multi-agent orchestration: chains, graphs, DAGs, and workflows.

prelude

Convenience re-exports for common Daimon types.

prompt

Composable prompt construction with variable interpolation.

retriever

Retrieval-Augmented Generation (RAG) abstractions.

server

HTTP server wrapper for exposing an agent as a REST API.

stream

Streaming response types for token-by-token or event-by-event model output.

telemetry

OpenTelemetry integration for exporting tracing spans to OTLP collectors.

tool

Tool abstraction and registry.

Enums

DaimonError

The central error type for all Daimon operations.

Type Aliases

Result

A type alias for Result<T, DaimonError>.

Attribute Macros

tool_fn

Derives a Tool implementation from an async function.