Warpcode for VS Code

A from-scratch, local coding agent that lives in your sidebar. Chat with an AI that can read and edit files, run commands, and search your code — without any managed runtime or closed-source agent framework.

Warpcode uses the OpenAI-compatible API, so it works with any compatible model provider: OpenAI, OpenRouter, DeepSeek, Groq, local vLLM/Ollama, and more. Project settings live in a committable .warpcode/warpcode.jsonc; API keys are stored separately in ~/.warpcode/auth.json (0600), shared with the Warpcode CLI.

Features

• 💬 Chat in the sidebar — ask questions, get explanations, write code
• 🛠️ Real tools — read/write/edit files, list files, grep, glob, and run bash commands
• 🔒 Permission prompts — approve or deny each file edit and command before it runs
• 📜 Session history — browse, resume, rename, and delete past conversations
• 🧭 Project navigation (v1.2) — find_file fuzzy filename search (frecency/mtime ranked), a project_overview summary injected into the first turn, and proactive lsp_diagnostics queries plus a per-turn workspace error section; all three toggles hot-switch from the settings page without restarting the extension
• ⚙️ Unified settings page — a single ⚙️ entrypoint with a left-nav layout: appearance, Providers, Credentials, MCP, LSP, feature toggles & permission mode, and a raw JSONC editor. Merged views show where each setting comes from (builtin/user/project) with live validation
• 🧵 Chronological timeline — text and tool calls display in the order they actually happen
• 🎨 Follows your VS Code theme
• 🔓 Open source (MIT) — read and modify every part

Setup

1. Install the extension
2. Open the Warpcode sidebar (the Warpcode icon in the right-side sidebar)
3. Click the ⚙️ Settings button
4. On the Providers tab, confirm your provider and set the default model (provider/model, e.g. openai/gpt-4o, deepseek/deepseek-chat, openrouter/openai/gpt-5)
5. On the Credentials tab, add an API key for your provider (writes ~/.warpcode/auth.json, 0600)
6. Start chatting

Built-in providers (openai, openrouter, deepseek, groq, ollama) need no endpoint config; custom OpenAI-compatible endpoints can be added by editing the JSONC config. Configuration is shared with the Warpcode CLI.

Usage

• Type a message and press Enter to send (Shift+Enter for a newline)
• Type / for slash commands:
  • /clear — clear the current conversation
  • /mode <name> — switch permission mode (default, acceptEdits, plan, bypassPermissions, dontAsk)
  • /help — show all commands
• Click + New to start a new session
• Click the 📜 History button to browse and resume past conversations
• File tools show an Open button to jump to the file; the bash tool shows a Terminal button for live output

Commands

• Warpcode: Open Chat — focus the sidebar
• Warpcode: Reload Session — restart the current session

Requirements

• VS Code 1.85 or newer
• An API key for an OpenAI-compatible model provider

Privacy

• Your code and conversations stay on your machine.
• API keys are stored locally in ~/.warpcode/auth.json (0600), never sent anywhere except your configured model endpoint. Project config (.warpcode/warpcode.jsonc) contains no secrets and is safe to commit.
• No telemetry, no accounts, no cloud.

License

MIT