Show HN: NanoAgent, zero-dependency 1k-LOC AI-agent runtime

a year ago

NanoAgent is a micro-framework for running LLM-powered agents in pure TypeScript with zero runtime dependencies outside of bun.
It focuses on the control loop, leaving RAG, vector search, databases, and cloud calls to MCP-compatible tools.
Key features include pure functions, deterministic stepping, built-in stuck detection, halting reasons, multi-stage workflows, and JSON-Schema tools.
NanoAgent supports first-class Bun, provides a tiny MCP server helper, and allows tools to live outside the agent process.
It enables RAG anywhere, horizontal scaling, and polyglot tooling without bloating the TS runtime.
The package is published as ES 2020 modules with type-definitions included and supports OpenAI and local models like Ollama.
Debugging is supported with a debug flag or custom logger.
Contributions are welcome under the MIT license, with a focus on maintaining no dependencies.

Hasty Briefsbeta