Trait-Based Plugins
Add LLM providers, vector stores, and task brokers by implementing traits. No lock-in, full composability.
Daimon is a high-performance Rust framework for AI agents. Trait-based plugins, zero-cost abstractions, async streaming, RAG, distributed execution, and multi-agent orchestration -- all compile-time safe.
No boilerplate. No ceremony. Build an agent, give it tools, and start prompting. Streaming, memory, and hooks are all opt-in.
Read the full guideuse daimon::prelude::*;
#[tokio::main]
async fn main() -> Result<(), DaimonError> {
let agent = Agent::builder()
.name("assistant")
.model(OpenAiModel::builder("gpt-4o").build()?)
.system_prompt("You are a helpful assistant.")
.build()?;
let response = agent.prompt("Hello, world!").await?;
println!("{}", response.text());
Ok(())
}Add LLM providers, vector stores, and task brokers by implementing traits. No lock-in, full composability.
Real-time token streaming with async/await. First-class cancellation, backpressure, and zero-copy design.
Built-in retrieval with pgvector, OpenSearch, Qdrant. Semantic search and document ingestion out of the box.
Supervisor, handoff, agent-as-tool, fork patterns. Chain, Graph, and DAG workflows.
Task brokers for Redis, NATS, RabbitMQ, SQS, Pub/Sub, and Service Bus. Scale horizontally.
Model Context Protocol client/server and Agent-to-Agent protocol. Full ecosystem interop.
Input/output validation, content policies, regex filters. Composable middleware pipeline.
Per-token cost tracking, budget limits, streaming cost events. Know exactly what you spend.
Save and resume agent runs. Time-travel debugging. Fork agents from any checkpoint.