ldclabs/anda

🤖 An AI agent framework built with Rust.

Anda – A Rust Framework for Composable AI Agent Runtimes

What it is

  • A Rust library that helps you build AI agents that can combine language models, tools, memory, and even other agents into a single, controllable runtime.
  • Emphasises type‑safe composition, asynchronous execution, and fine‑grained runtime control.

Core ideas

  1. Composable agents & tools – Define agents and tools via stable traits; plug them together without hard‑coding the workflow.
  2. Model routing – Tag model instances (primary, pro, flash, lite) and let the engine automatically pick the right one for a request.
  3. Runtime orchestration – CompletionRunner manages the whole turn: model calls, tool invocations, usage accounting, cancellation, and hand‑offs for long sessions.
  4. Scoped execution context – BaseCtx / AgentCtx give each agent its own isolated state, cache, HTTP client, and cancellation token.
  5. Extensible memory & skills – Optional modules add conversation storage, KIP‑based memory tools, filesystem access, shell execution, notes, todos, and file‑backed skill catalogs.
  6. Discovery‑aware tool bundles – Agents can query tool groups (tools_groups) and load only the schemas they need (tools_select).

Project layout

  • anda_cli – CLI to start and interact with an engine.
  • anda_core – Core traits, types, and contracts that developers implement.
  • anda_engine – The runtime: orchestration, contexts, model adapters, extensions.
  • anda_engine_server – HTTP server exposing one or more engines.
  • anda_web3_client – Web3 client for environments without a trusted execution environment.

How to get started

  1. Clone the repo and run cargo build.
  2. Use anda_cli to launch a server: anda_cli serve --config my_engine.toml.
  3. Implement custom agents/tools by implementing the traits in anda_core and register them with the engine.
  4. Interact via the HTTP API or the CLI to send prompts, invoke tools, or manage memory.

Who might use it

  • Developers building personal assistants, chat‑bots, or autonomous agents that need to call external tools.
  • Teams that want a Rust‑native alternative to Python‑based agent frameworks, with strong type safety and async performance.
  • Projects that require persistent memory or knowledge‑graph integration (via the KIP protocol).

Related products

  • Anda Brain – A persistent memory and cognition service built on top of this framework.
  • Anda Bot – A ready‑made personal AI assistant using the same runtime.

License

  • MIT License (© 2026 LDC Labs).

Anda is a genuine, actively maintained open‑source project aimed at developers who need a robust, composable runtime for AI agents in Rust.

Related

  • Project
    liquidos-ai/AutoAgents

    A production-grade multi-agent framework in Rust that provides a type-safe environment for building intelligent systems with structured tool calling and pluggable LLM backends.

  • Project
    zavora-ai/adk-rust

    ADK‑Rust is a production‑ready Rust framework (43 crates) for building, orchestrating and serving AI agents. It is model‑agnostic (Gemini, OpenAI, Anthropic, Ollama, Bedrock, local mistral.rs, etc.), offers composable feature tiers (minimal → full), supports durable graph workflows, tool calling, RAG, realtime voice/video, payments, and provides an embedded web UI for monitoring agent execution.

  • Project
    eastreams/loong

    A Rust-based framework for building secure and extensible vertical AI agents capable of long-horizon workflows and compound task execution.

  • Project
    Dicklesworthstone/pi_agent_rust

    pi_agent_rust is a native Rust CLI AI coding assistant that streams LLM responses, persists sessions in SQLite, and offers a capability‑gated extension system. It runs as a single binary, supports local (Ollama, llama.cpp, etc.) and OpenAI‑compatible providers, and emphasizes fast startup, low memory use, and strong security controls.

  • Project
    Abraxas-365/langchain-rust

    A Rust implementation of the LangChain framework for building composable LLM applications with support for various models, vector stores, and agents.