SmallCoder
An open-source, free AI coding agent tool optimized for locally deployed open-source models. It adapts to smaller models through streamlined system prompts and context compression mechanisms, and supports automatic detection of models deployed via Ollama and LM Studio.
Timeline (last 90 days)
SmallCoder is an open-source, free local AI programming agent tool that can be installed via npm install -g smallcoder and has been optimized for local open-source models from the system prompt and toolset to the compression mechanism.
SmallCoder automatically detects local models deployed via Ollama or LM Studio when selecting a model, requiring no additional configuration.
SmallCoder provides both terminal and web interfaces; the terminal is launched with the small command, and the web interface is launched with small --web and generates a standalone URL.
SmallCoder allows switching between three permission modes—read-only, edit, and bypass—by holding Shift and pressing Tab.
SmallCoder's planning mechanism writes task progress to a separate document, enabling the agent to review the plan, confirm completed tasks, and stay on track after each context compression.
SmallCoder has no built-in browser engine but provides a built-in browser view where a local URL can be pasted to view running results.
SmallCoder's visual error-correction feature depends on the model itself supporting it; Qwen3 supports vision, tool calling, and thinking tags simultaneously.
The author believes small models will never reach the level of flagship models, and the value of SmallCoder lies in improving the practical usability of local small models rather than making them stronger.
The author used SmallCoder with a local model to complete a Minecraft clone project based on Three.js, with the build taking approximately 20 minutes and completing without interruption.
All Facts (9)
SmallCoder is an open-source, free local AI programming agent tool that can be installed via npm install -g smallcoder and has been optimized for local open-source models from the system prompt and toolset to the compression mechanism.
50%UnverifiedSmallCoder automatically detects local models deployed via Ollama or LM Studio when selecting a model, requiring no additional configuration.
50%UnverifiedSmallCoder provides both terminal and web interfaces; the terminal is launched with the small command, and the web interface is launched with small --web and generates a standalone URL.
50%UnverifiedSmallCoder allows switching between three permission modes—read-only, edit, and bypass—by holding Shift and pressing Tab.
50%UnverifiedSmallCoder's planning mechanism writes task progress to a separate document, enabling the agent to review the plan, confirm completed tasks, and stay on track after each context compression.
50%UnverifiedSmallCoder has no built-in browser engine but provides a built-in browser view where a local URL can be pasted to view running results.
50%UnverifiedSmallCoder's visual error-correction feature depends on the model itself supporting it; Qwen3 supports vision, tool calling, and thinking tags simultaneously.
50%UnverifiedThe author believes small models will never reach the level of flagship models, and the value of SmallCoder lies in improving the practical usability of local small models rather than making them stronger.
50%UnverifiedThe author used SmallCoder with a local model to complete a Minecraft clone project based on Three.js, with the build taking approximately 20 minutes and completing without interruption.
50%