LocalAI Code Editor

Free, open-source, self-hosted AI code editor powered by local LLMs
_{No cloud. No API keys. No telemetry. Your code stays on your machine.}

A free, open-source, self-hosted AI code editor powered entirely by your local LLM — no cloud, no API keys, no telemetry.

What is LocalAI Code Editor? LocalAI Code Editor is a free, open-source, self-hosted AI code editor that runs entirely on your local machine. It connects to local LLMs — Ollama, LM Studio, llama.cpp, vLLM, or any OpenAI-compatible server — to provide an agentic AI coding assistant without sending your code to the cloud. Think VS Code + AI, but private.

Key features: Monaco Editor (same as VS Code), agentic AI that reads/writes files and runs commands, MCP protocol support for external tools, a marketplace of 50+ agent skills, Docker-in-Docker sandboxing, LSP IntelliSense, full git panel with blame, and builds for Windows, macOS, Linux, Docker, and Unraid.

🎯 Why LocalAI Code Editor?

Feature	Benefit
🔒 100% Private & Offline	Every completion, refactor, and agent run uses your local models. No data ever leaves your machine or your network.
🛠️ Full Agentic Assistant	The AI reads/writes files, runs shell commands, calls MCP tools, loads project skills, and iterates until the job is done.
🔌 MCP-First Architecture	Connect any local (stdio) or remote (HTTP/SSE) MCP server — its tools become available to the agent automatically.
🧠 Agent Skills System	Drop SKILL.md files into your project or home directory to teach the agent your conventions, workflows, and best practices.
📚 Real IntelliSense	LSP language servers provide completion, hover, diagnostics, go-to-definition, find-references, and rename.
📋 Complete Git Panel	Status, diff, stage/unstage, commit, branch management, push/pull, and log — all without leaving the editor.
🚀 One Codebase, Every Target	Runs as a desktop app (Electron), local web app, or Docker container (Unraid, Kubernetes, any Docker host).

📋 Table of Contents

• Features
• Supported LLM Providers
• Quick Start
• Configuration
• Agent Skills
• Language Servers (LSP)
• MCP Servers
• Docker & Unraid Deployment
• Development
• Project Structure
• Roadmap
• Changelog
• Contributing
• Security
• License

---

✨ Features

Category	Capabilities
AI Agent	Streamed chat that reads/writes files, runs commands, calls MCP tools, and iterates to completion
Local LLM Support	Auto-detect Ollama, LM Studio, llama.cpp, and any OpenAI-compatible endpoint
MCP Client	Connect local (stdio) and remote (HTTP/SSE) MCP servers; expose their tools to the agent
Agent Skills	50+ built-in skills; project + user SKILL.md files with frontmatter; loaded on demand
Skills Marketplace	Installed/Discover tabs with 18 curated skills across 12 categories; one-click install
Language Servers	LSP-based completion, hover, diagnostics, definition, references, rename
Git Panel	Status, diff, stage/unstage, commit, stash, branch manager, push/pull, log with diff preview, blame
File Explorer	Browse, open, edit, save — plus New File/Folder, Rename, and Delete from the sidebar
Quick Open & Search	Ctrl+P fuzzy file finder; workspace-wide text search with regex; @file mentions
Monaco Editor	The industry-standard editor used by VS Code, running locally

View all 28 features…

Category	Capabilities
Agent Task History	View, replay, and branch from previous agent task trajectories
Model Comparison	Run the same prompt against multiple models side-by-side
Agent System Prompt Tuning	Preset templates, character count, expanded editor, template variables ({{workspace}}, {{git_branch}}, etc.)
Approve-before-Apply	Approve/deny per write+command via Agent panel toggle (diff-level review)
Docker-in-Docker Sandbox	Isolated Docker container for safe agent code execution — start/stop/exec via Settings
Provider Setup UI	Add, edit, test, and remove providers at runtime (Settings panel) — persisted per workspace
Workspace Management	Native folder picker on launch; switch projects anytime; live-rebinds git/skills/LSP
MCP Server Discovery	Curated registry of 15+ servers with search, category filter, and one-click install
MCP Tool Discovery	Searchable tool list grouped by server with expandable sections and descriptions
Token Usage Dashboard	Visual input/output breakdown, cost estimates, locale-formatted token counts
Conversation Export	Export chat history as Markdown or JSON for sharing and debugging
Skill Auto-Application	Workspace-aware skill suggestions based on project indicators and path heuristics
Terminal Integration	Run commands in an integrated terminal over WebSocket
Code Formatting	Format code with language-specific rules; auto-save with configurable delay
Command Palette	Quick access to all editor commands via Ctrl+Shift+P
Settings Persistence	Editor preferences saved to localStorage; workspace settings in .localai/
Multi-tab Support	Tab management utilities for multiple files
Error Handling	Structured error types and retry logic throughout
Keyboard Shortcuts	Reference for all editor commands with platform-specific keys
File Change Tracking	Track agent modifications to files with change types
Editor Settings UI	Auto-save toggle, minimap visibility, font size, word wrap controls
Markdown Preview	Toggle Edit/Preview for .md files
Multi-Platform	Windows, macOS, Linux (Electron), and any web-capable device (Docker)

---

🤖 Supported LLM Providers

LocalAI Code Editor connects to every major local inference server through their OpenAI-compatible APIs:

• Ollama — http://localhost:11434/v1
• LM Studio — http://localhost:1234/v1
• llama.cpp server — http://localhost:8080/v1
• vLLM — http://localhost:8000/v1
• Jan — http://localhost:1337/v1
• Text-Generation-WebUI — http://localhost:5000/v1
• Any OpenAI-compatible endpoint — Custom servers, enterprise proxies, and more

The editor health-checks each endpoint and lets you pick the running provider and model from the Agent panel.

Configuring providers from the UI

No config file needed — open the Settings panel (⚡ icon) → LLM Providers:

• See every provider with live health/latency
• Add a provider from a preset (Ollama, LM Studio, llama.cpp, vLLM) or any custom OpenAI-compatible URL
• Edit label / base URL / API key, Test the connection before saving, or Remove
• Settings persist to <workspace>/.localai/settings.json and survive restarts (dev, desktop, and Docker)

---

🚀 Quick Start

Requirements

• Node.js 20+ and pnpm 10
• A local LLM server (e.g. Ollama) running with at least one model pulled:

  ollama pull codellama:13b
  ollama serve

Install & Run (Web UI)

git clone https://github.com/wildfirebill-ai/localai-code-editor.git
cd localai-code-editor
pnpm install
pnpm build
pnpm dev:server        # starts the server + web UI at http://127.0.0.1:4801

Open http://127.0.0.1:4801 in your browser, pick your provider and model in the Agent panel, and start asking the AI to build, edit, run, and debug your code.

Desktop App (Electron)

pnpm --filter @localai/desktop dev        # run with hot reload
pnpm --filter @localai/desktop dist       # package installers for your OS

On first launch the desktop app asks you to pick a project folder (native dialog). You can switch projects anytime from Settings → Workspace → Open Folder…, or by typing an absolute path (also works in browser/Docker mode).

Or skip the build entirely — grab a ready-made installer from the latest release.

Docker (Production / Unraid)

docker run -d --name localai-code-editor \
  -p 4801:4801 \
  -v /path/to/your/repo:/workspace \
  -v /path/to/config:/root/.localai \
  ghcr.io/wildfirebill-ai/localai-code-editor:latest

• WebUI: http://<your-host>:4801
• /workspace = the repo you want to edit
• /root/.localai = persistent skills + config

---

⚙️ Configuration

Create a localai.config.json in your workspace (full example: localai.config.json.example):

{
  "workspace": "/path/to/your/repo",
  "port": 4801,
  "host": "127.0.0.1",
  "providers": [
    { "id": "ollama", "label": "Ollama", "baseUrl": "http://localhost:11434/v1" }
  ],
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/workspace"]
    }
  },
  "languageServers": [
    { "id": "typescript", "language": "typescript", "extensions": [".ts", ".tsx"], "command": "typescript-language-server", "args": ["--stdio"] }
  ],
  "protectedPaths": [".git"],
  "allowShell": true
}

Key	Purpose
providers	LLM endpoints to connect to (auto-detects Ollama, LM Studio, etc.)
mcpServers	MCP servers to launch at startup (stdio or HTTP)
languageServers	LSP servers to manage for IntelliSense
allowShell	Whether the agent may run shell commands
protectedPaths	Paths the file tools may not touch (e.g. .git, .env)

---

🧠 Agent Skills

50+ production-grade skills ship built-in and work in every workspace — commit, test-and-fix, typecheck-fix, lint-clean, debug-failure, refactor-safe, docker-build-run, dependency-update, git-release, write-docs, security-check, and 40 more covering git, quality, testing, performance, frontend, backend, ops, and AI integration.

Skills Marketplace

The Skills panel (🧠 icon) has two tabs:

• Installed — all loaded skills with enable/disable controls and workspace-aware suggestions
• Discover — browse 18 curated skills across 12 categories (git, backend, frontend, devops, testing, security, quality, docs, docker, ai-integration, performance, maintenance) with search and one-click install

Custom Skills

Skills are Markdown files with frontmatter that teach the AI your conventions, workflows, and best practices. Drop a SKILL.md into either location:

• Project skills: <workspace>/.localai/skills/<name>/SKILL.md (overrides builtins)
• User (global) skills: ~/.localai/skills/<name>/SKILL.md

---
name: ts-check
description: Typecheck a package with pnpm typecheck before committing
category: quality
---
Run `pnpm typecheck` from the package root and fix any errors.

Project skills override same-named global skills. The agent loads them on demand via the read_skill tool.

---

📚 Language Servers (LSP)

Add language servers to languageServers in your config for full IntelliSense:

{
  "languageServers": [
    { "id": "typescript", "language": "typescript", "extensions": [".ts", ".tsx", ".js", ".jsx"], "command": "typescript-language-server", "args": ["--stdio"] },
    { "id": "python", "language": "python", "extensions": [".py"], "command": "pyright-langserver", "args": ["--stdio"] },
    { "id": "rust", "language": "rust", "extensions": [".rs"], "command": "rust-analyzer", "args": [] }
  ]
}

Install the server binaries yourself:

npm i -g typescript-language-server pyright rust-analyzer

Features: completion, hover, diagnostics, go-to-definition, find-references, rename.

---

🔌 MCP Servers

Connect any MCP server — local subprocesses and remote endpoints:

{
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/workspace"]
    },
    "github": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"]
    },
    "postgres": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://user:pass@localhost/db"]
    },
    "remote": {
      "type": "http",
      "url": "http://localhost:9000/mcp"
    }
  }
}

MCP Server Discovery

The MCP Servers panel has two tabs:

• Installed — connected servers with status indicators, tool counts, and a manual add form
• Discover — curated registry of 15+ servers (Filesystem, Git, GitHub, PostgreSQL, SQLite, Redis, Web Fetch, Brave Search, AWS, Kubernetes, Hugging Face, Memory, Sequential Thinking) with search, category filter, and one-click install

MCP Tool Discovery

Connected tools are searchable and grouped by server in Settings → Connected Tools. Hover over any tool to see its description.

---

🐳 Docker & Unraid Deployment

LocalAI Code Editor runs headless in Docker and is ideal for Unraid via Community Apps or a custom template. The image is published multi-arch (linux/amd64, linux/arm64) to GHCR by CI.

docker run -d --name localai-code-editor \
  -p 4801:4801 \
  -v /mnt/user/your_repo:/workspace \
  -v /mnt/user/appdata/localai:/root/.localai \
  ghcr.io/wildfirebill-ai/localai-code-editor:latest

• WebUI: http://<your-unraid-ip>:4801
• /workspace = the repo you want to edit
• /root/.localai = persistent skills + config

⚠️ Docker Socket Warning

If you want the editor to build and test Docker programs from inside the editor, you must mount the Docker socket:

volumes:
  - /var/run/docker.sock:/var/run/docker.sock

WARNING — read this before mounting the Docker socket.

Mounting /var/run/docker.sock into a container grants root-equivalent control over the entire Docker host to that container. Because LocalAI Code Editor's AI agent can run arbitrary shell commands, anything that can reach the editor's web UI — including a malicious prompt, an untrusted repository it opens, or any user who can reach port 4801 — could effectively take over your Unraid server and everything else on the Docker host.

Use the socket mount only on a trusted, single-user server where you alone control access. Never expose the editor's UI to the internet or to untrusted users while the socket is mounted. Prefer to keep it on your LAN only.

Alternative Deployment Options

Option	What it does	Security
B — Edit here, build on the host	Develop the Docker program in the editor; build/run it on the Unraid host against the same mounted /workspace repo.	✅ No socket, no risk. Recommended default.
C — Docker-in-Docker (DinD)	Run a nested Docker daemon inside the editor container, isolated from the host.	✅ Isolated from the host. Heavier (privileged container).
A — Docker CLI + socket	Mount the socket + docker CLI so the agent can run docker build / docker compose directly.	⚠️ Root-equivalent host access. Only for trusted single-user servers.

In short: start with Option B for safety. Upgrade to Option A only if you truly need the agent to drive Docker from inside the editor and you fully control the network.

---

🛠️ Development

# Setup
pnpm install

# Development servers
pnpm dev:server   # backend on :4801 (also serves the web UI)
pnpm dev:web      # Vite dev server with hot reload on :5173
pnpm dev:desktop  # Electron with hot reload

# Build & Quality
pnpm build        # build all packages
pnpm typecheck    # typecheck all packages
pnpm lint         # lint all packages
pnpm test         # run tests

See Project Structure and CONTRIBUTING.md for details.

---

📁 Project Structure

packages/
  provider/   LLM adapters — Ollama, LM Studio, llama.cpp, OpenAI-compatible
  mcp/        MCP host — local (stdio) + remote (HTTP/SSE) servers, tool calls
  agent/      Agentic loop — shell/file tools, tool-call iteration
  git/        Git panel engine — status, diff, branches, push/pull
  skills/     SKILL.md loader — project + user, frontmatter parsing
  lsp/        Language-server host — spawn + JSON-RPC over stdio, WS bridge
  server/     Node backend — WebSocket JSON-RPC + static web UI
apps/
  web/        Editor UI — Monaco, explorer, agent/chat, git panel, MCP/skills
  desktop/    Electron shell — spawns the server, loads the UI

---

🗺️ Roadmap

See ROADMAP.md for the detailed roadmap.

High-Level Milestones

Milestone	Target	Status
Core editor + agent loop	v0.1	✅ Done
MCP + Skills system	v0.1	✅ Done
LSP IntelliSense	v0.1	✅ Done
Git panel + blame	v0.1–v0.2	✅ Done
Multi-platform (Win/Mac/Linux/Docker)	v0.1	✅ Done
Agent Skills Marketplace	v0.2	✅ Done
Docker-in-Docker Sandbox	v0.2	✅ Done
Agent System Prompt Tuning	v0.2	✅ Done
Skill Auto-Application	v0.2	✅ Done
Enhanced MCP Tool Discovery	v0.2	✅ Done
Editor Polish (themes, keybindings, navigation)	v0.3	📋 Planned
Collaboration & Scale (multi-user, sessions)	v0.4	📋 Planned
Extensibility Platform (plugins, extensions)	v0.5	📋 Planned
Stability & Polish (tests, a11y, i18n)	v1.0	📋 Planned

---

🏷️ CI & Releases

Both workflows (ci.yml, docker.yml) run only when:

1. A version tag is pushed — git tag v0.1.4 && git push --tags — or
2. Manually triggered from the Actions tab (workflow_dispatch)

Everyday pushes to main and pull requests do not run CI. Pushing a v* tag triggers the full pipeline:

Job	What it does
test	Builds + typechecks + lints + tests on Windows, macOS, and Linux
desktop	Packages installers per OS (NSIS/zip, DMG/zip, AppImage/deb/tar.gz)
release	Publishes all desktop artifacts to a GitHub Release with generated notes
docker	Builds and pushes the multi-arch image to ghcr.io with semver tags

---

📜 Changelog

See CHANGELOG.md for detailed release notes.

Recent Releases

Version	Date	Highlights
v0.2.7	2026-08-26	Theme Switcher, Release Notes Modal, Skill Versioning, Skill Search
v0.2.6	2026-08-26	Find & Replace, Debug Console, Bracket Matching, Multi-cursor, Code Folding, MCP Resilience (includes v0.2.4 + v0.2.5)
v0.2.3	2026-08-25	Git Stash, Interactive Commit Editor, Commit History Viewer, Branch Comparison
v0.2.2	2026-08-25	Prompt Variables, Chat Export, Tool Description Customization, Token Usage Dashboard
v0.2.1	2026-08-25	Skills Marketplace, Docker-in-Docker Sandbox, Prompt Tuning UI, Skill Suggestions, Enhanced Tool Discovery
v0.2.0	2026-08-23	MCP Discovery, Task History, Model Comparison, Terminal, Git Blame, Code Formatting, Settings UI
v0.1.6	2026-08-22	Electron 43 (clears flagged CVEs), Docker slimmed + hardened, automated vuln scanning
v0.1.5	2026-08-21	Quick Open (Ctrl+P), file search, @file mentions, 51 builtin skills, md preview
v0.1.0	2026-08-20	Initial release: Core editor, agent loop, MCP, Skills, LSP, Git panel, Win/Mac/Linux/Docker

---

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

Quick Contribution Guide

1. Fork the repo
2. Create a feature branch: git checkout -b feat/amazing-feature
3. Make your changes with tests
4. Run quality checks: pnpm typecheck && pnpm lint && pnpm test
5. Submit a PR with a clear description

---

🔒 Security

See SECURITY.md for our security policy and how to report vulnerabilities. For full transparency, VULNERABILITIES.md documents every known weakness, attack surface, and accepted design tradeoff — including prompt-injection risk and the unauthenticated server — before you discover them the hard way.

Key Security Principles

• No external network calls unless you configure a remote provider/MCP
• Shell commands are opt-in (allowShell: true in config)
• Protected paths prevent the agent from touching sensitive files
• Docker socket mount is opt-in with explicit warnings

To report a security issue, email security@wildfirebill.ai or use GitHub Security Advisories.

---

📄 License

MIT License — free to use, modify, and self-host.

---

🙏 Acknowledgments

• Monaco Editor — The editor that powers VS Code
• Model Context Protocol — For the MCP specification
• Ollama — Making local LLMs accessible
• Electron — Cross-platform desktop apps
• TypeScript — Type-safe JavaScript at scale

---

Star ⭐ this repo if you find it useful! It helps others discover LocalAI Code Editor.

---