little-coder
A harness optimized to smaller LLMs
little-coder is an AI coding agent specifically tuned for small local language models (particularly Qwen 9.7B–35B), built on top of the pi framework. It achieves competitive performance on coding benchmarks through architectural adaptations including skill injection, knowledge cheat-sheets, output parsing, and quality monitoring tailored to smaller models. The agent ships with 30+ extensions and 30 skill markdown files that optimize scaffold-model fit, enabling local models to outperform some frontier models on benchmarks like Aider Polyglot. It operates as a standalone terminal-based coding assistant that reads, writes, edits files, executes shell commands, and manages background jobs in your project directory.
Key Features
Use Cases
- 01Running a coding agent entirely on local hardware with 8GB VRAM consumer laptops
- 02Planning complex features with a large model then implementing with a faster small model
- 03Managing long-running background tasks (training, builds, servers) with intelligent event-based interrupts
- 04Conducting deep research on external topics with isolated sub-agents and consolidated reports
- 05Operating on large markdown knowledge bases with citation and cross-reference management
- 06Benchmarking small model performance on Aider Polyglot, Terminal-Bench, and GAIA datasets
Related Agents
View morehermes-agent
The agent that grows with you
agency-agents
A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.
openinterpreter
A coding agent for open models like Kimi K3
cline
Autonomous coding agent as an SDK, IDE extension, or CLI assistant.
little-coder — FAQ
What is little-coder and what does it do?+
little-coder is a terminal-based AI coding agent optimized for small local language models (9.7B–35B parameters). It provides architectural scaffolding—skill injection, output parsing, quality monitoring—that enables smaller models to perform coding tasks competitively with larger models through better scaffold-model fit.
How do I install little-coder?+
Install globally via npm with 'npm install -g little-coder' (requires Node.js 22.19+). Optionally set up a local llama.cpp, Ollama, or LM Studio server for local inference, or configure cloud provider API keys (Anthropic, OpenAI) for cloud models.
What models and providers does it support?+
It works with local providers (llama.cpp, Ollama, LM Studio) serving models like Qwen3.5-9B or Qwen3.6-35B-A3B, and cloud providers (Anthropic Claude, OpenAI GPT-4o-mini). The default configuration targets local llama.cpp serving Qwen3.6-35B-A3B on port 8888.
Do I need API keys or special prerequisites?+
For local providers, set a placeholder API key (e.g., 'export LLAMACPP_API_KEY=noop'). For cloud providers, set the standard environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY). You'll also need Node.js 22.19+ installed.
Is little-coder free to use?+
Yes, little-coder itself is open source (Apache 2.0). When using local models, there are no usage costs beyond your hardware. Cloud model usage incurs the provider's standard API fees.
Which AI clients or frameworks does it work with?+
little-coder is a standalone agent built on the pi framework, launched from the terminal with the 'little-coder' command. It's not a plugin for other AI clients, though community bridges exist for integrations like Zed's ACP.
How do I install little-coder?+
Open the source repository on GitHub and follow its README. little-coder is a agent — MCP Agents Market links you directly to the official repo.
Is little-coder free?+
little-coder is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.