About this project

LM Studio Code is a Visual Studio Code extension that provides a full coding-agent panel instead of simple autocomplete. It runs against models served by LM Studio on the user's own machine, so no cloud API key, token billing, or internet round-trip is required. The agent is powered by the open-source OpenCode agent, which is bundled with the extension and can work offline. The agent can read and edit workspace files, run shell commands, search the codebase, and display each action as an expandable tool card. Users remain in control through inline permission prompts and a permissions picker with Auto, Manual, and Bypass modes. It offers build and plan modes, a live todo list, streaming Markdown responses, collapsible reasoning blocks for supported models, and adjustable reasoning effort where the model exposes it. Advanced features include custom agents defined by Markdown files in .opencode/agent, subagent delegation with separate context windows, and autonomous goals started with /goal. Goals are checked by a local judge model, tracked in a pinned goal bar, capped at 25 rounds, and protected by stall detection. Users can also inject new instructions while the agent is working. The extension integrates tightly with LM Studio: it lists loaded and unloaded models with context size, publisher, format, and quantization; can reload models with adequate context via the lms CLI; supports multiple local or remote LM Studio servers; and stores remote API keys in VS Code encrypted Secret Storage. Context tools include attaching selected code, adding the open file, pasting images, compacting conversations, and a context meter. Sessions are restored automatically, can be renamed or resumed, and may be opened in editor tabs for parallel work. It also supports MCP servers and skills, automatically discovering .mcp.json and .vscode/mcp.json configurations, with /mcp status reporting and slash commands such as /clear, /compact, /file, /skills, and /help. Requirements are VS Code 1.104+, a running LM Studio local server, and preferably the lms CLI for context management.