llama.vscode
https://github.com/ggml-org/llama.vscode- Category
- Developer Tools
- Rank
- No. 609Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- ggml-org
- GitHub
- 1.5k stars
- Latest release
- v0.0.65
- Date
About
VS Code extension for local LLM-assisted code and text completion, powered by llama.cpp.
What it does
It brings inline suggestions, chat, text editing, commit-message generation, and an agent into VS Code. You choose grouped model environments for completion, chat, embeddings, and tools, using local servers or configured external endpoints.
Why it's ranked here
The appeal is breadth with meaningful local control. Partial suggestion acceptance, reusable context, model grouping, performance statistics, and agent tools cover more than basic autocomplete. That breadth also brings substantial model and hardware setup.
What's good
You can accept a whole suggestion, one line, or one word. Context can include nearby text, open and edited files, and copied material. The agent can use installed MCP servers, while model environments make coordinated selection easier.
Tradeoffs
Completion requires a fill-in-the-middle compatible model. Local quality and speed depend on available hardware, and the documented CPU-only choices produce significantly lower quality. Linux installation requires obtaining binaries and adding them to the system path.
How to use it well
It suits VS Code users who want configurable local completion plus optional chat and coding agents. Start with a predefined environment matched to your hardware, then tune context and generation time. It does not provide the documented Vim or Neovim integration itself.
Technical notes+
package.json declares a VS Code extension requiring VS Code ^1.109.0, with inline-completion, chat, editing, agent, and commit-message commands plus configurable endpoints and local launch commands. src/lists.ts defines predefined local and external models across completion, chat, embeddings, tools, and grouped environments. src/tools.ts registers workspace search, file access, editing, terminal, diagnostics, symbol-renaming, delegation, and VS Code MCP tool support; terminal execution asks permission when commands may modify state. src/git.ts reads staged changes first, falls back to unstaged changes, and writes generated text into the first repository’s commit input. src/utils.ts contains BM25 scoring and command classification. src/types.ts models environments, agents, chats, model endpoints, and indexed chunks.
Observed
- Primary language
- TypeScript
- Packaging
- VS Code extension published through the VS Code Marketplace and Open VSX
- Interface
- VS Code commands, keybindings, inline suggestions, menus, and an Explorer webview
- Model interfaces
- Configurable endpoints and local server launch commands for completion, chat, embeddings, and tools
- Platform support
- Documented setup for macOS, Windows, and Linux
- Agent integration
- Can select tools from MCP servers installed and started in VS Code
Read from README.md, package.json, src/git.ts, src/menu.ts, src/lists.ts, src/tools.ts, src/types.ts, src/utils.ts, src/logger.ts, src/plugin.ts, src/dialogs.ts, src/prompts.ts.
What it can do
Generate code completions
Partial code or programming context → Code suggestions and completions
Complete text content
Partial text or writing context → Text suggestions and completions
Generate code from natural language prompts
Natural language description of desired code → Generated code snippets
Provide code suggestions in real-time
Active code editing session → Contextual code recommendations
Process local LLM inference
Code or text prompts → AI-generated responses via llama.cpp
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.