microagents-core

Core agent runtime for the MicroAgents ecosystem. It provides the building blocks for creating LLM-powered agents that can use tools, load domain-specific skills, and persist conversation sessions across multiple storage backends.

What it does

• Agent construction — configure an agent via a fluent builder, choosing the LLM provider, model, tools, skills, and storage backend.
• Streaming generation — talk to OpenAI, Groq, OpenRouter, or a local Ollama instance and receive token-level streaming responses.
• Tool calling — register Rust implementations of ToolFunction; the runtime validates JSON arguments against a schema, executes tools (optionally in parallel), and feeds results back to the model.
• Skills — load markdown-based skill definitions from .agents/skills or ~/.agents/skills. Skills are injected into the system prompt so the model can reason about specialised domains.
• Session persistence — resume conversations from JSONL, SQLite, or in-memory storage.

Quick example

use microagents_core::agent::MicroAgentBuilder;
use microagents_core::types::ToolExecutionContext;
use futures_util::StreamExt;

let agent = MicroAgentBuilder::new(ToolExecutionContext::new(()))
    .provider("openai".into())?
    .model("gpt-5.5".into())
    .find_skills()?
    .build()?;

let mut stream = agent.run("Hello!".into(), None).await?;
while let Some(event) = stream.next().await {
    println!("{:?}", event?);
}

Module overview

Supported providers

• OpenAI — requires OPENAI_API_KEY
• Groq — requires GROQ_API_KEY
• OpenRouter — requires OPENROUTER_API_KEY
• Ollama — reads OLLAMA_BASE_URL (defaults to http://localhost:11434/v1)