A minimalist, terminal-native coding agent written in C that works with local llama.cpp models and many cloud LLM providers.
itayinbarr/little-coder
A coding agent built on pi, tuned for small local LLMs, with background jobs, plan mode and per-phase model selection.
Open-source alternative to:Claude Code
- Stars
- 2,718
- Forks
- 187
- Open issues
- 11
- Last commit
- 2 months ago
What is it?
little-coder is a terminal coding agent designed to get useful work out of small local language models rather than frontier cloud models. It is built on top of the pi agent framework and adds about thirty extensions, thirty skill cards that inject short tool-usage guidance, and a Python benchmark harness. The launcher loads only the bundled extensions, which keeps the cold-start context small and behavior predictable.
The agent works against llama.cpp, Ollama, LM Studio and any OpenAI-compatible server, and it also supports cloud providers such as Anthropic and OpenAI. Features include a plan mode that researches a task and writes a plan before editing, a deep research mode that produces a cited report, sub-agents spawned through a dispatch tool, and background shell jobs that wake the model on matches, silence or exit events. Shell commands pass through a safe-prefix whitelist, and per-phase model settings let planning and implementation run on different models.
The project publishes Aider Polyglot and Terminal-Bench results for Qwen models run on a single consumer laptop GPU. It is installed with npm or bun and requires Node.js 22.19 or newer.
Key features
- 01Runs on local llama.cpp, Ollama and LM Studio servers as well as cloud providers
- 02Plan mode researches a task, asks clarifying questions and writes a plan before editing
- 03Deep research mode produces a single cited markdown report
- 04Background shell jobs wake the model on output patterns, silence or exit
- 05Shell commands are gated by a safe-prefix whitelist and read-before-edit rules
- 06Separate models can be assigned to planning and implementation phases
Use cases
- Running a coding agent offline on a laptop with a locally hosted Qwen or similar model
- Comparing how small models perform on coding benchmarks with a fixed scaffold
- Running long builds or training jobs while the agent monitors them in the background
- Planning a change with a large model, then implementing it with a small local one
Quick start
- 1
Install Node.js 22.19 or newer
- 2
Install little-coder globally with npm
npm install -g little-coder - 3
Run little-coder inside your project directory
little-coder
Things to know
- Designed and tuned around local llama.cpp setups with Qwen models, so cloud models get less attention.
- Requires Node.js 22.19 or newer at runtime even when installed with bun.
- Running the recommended local model needs a capable GPU or substantial RAM and a manual llama.cpp build.
In collections
Topics
FAQ
- What is little-coder?
- A coding agent built on pi, tuned for small local LLMs, with background jobs, plan mode and per-phase model selection. little-coder is a terminal coding agent designed to get useful work out of small local language models rather than frontier cloud models. It is built on top of the pi agent framework and adds about thirty extensions, thirty skill cards that inject short tool-usage guidance, and a Python benchmark harness. The launcher loads only the bundled extensions, which keeps the cold-start context small and behavior predictable.
- Is little-coder free to use? What license does it have?
- little-coder is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is little-coder still maintained?
- Updated within 6 months. The last commit to little-coder was on Sep 18, 2026.
- What can I use little-coder for?
- Running a coding agent offline on a laptop with a locally hosted Qwen or similar model Comparing how small models perform on coding benchmarks with a fixed scaffold Running long builds or training jobs while the agent monitors them in the background Planning a change with a large model, then implementing it with a small local one
- How do I install little-coder?
- Install Node.js 22.19 or newer Install little-coder globally with npm: npm install -g little-coder Run little-coder inside your project directory: little-coder
- Where can I get little-coder?
- The source code and installation instructions are on GitHub: https://github.com/itayinbarr/little-coder
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A desktop coding agent (Tauri app plus CLI) that runs on your existing AI subscriptions or local models, with built-in code search and verification checks.
Pi, the AI coding agent, compiled ahead of time to native code so it starts faster and uses less CPU than Pi on Bun or Node.
An open-source coding agent that reads your project, edits files and runs tests from your terminal, desktop, browser or editor.
A local-LLM-first agentic coding assistant that runs against vLLM, sglang, ollama or llamacpp backends with a browser UI.
VS Code extension for local LLM code completion, AI chat, and an agentic coding assistant backed by llama.cpp.