TeleCode AI

Remote-controlled autonomous coding agent in your VS Code.
Send a task from your phone. TeleCode edits your project, shows the changes, and sends the result back.

TeleCode AI runs coding tasks in your VS Code workspace, locally or from Telegram and WhatsApp. Ask it to investigate a bug, edit code, or run checks. Follow progress from your phone, inspect the diff, and commit the result when you are ready.

/run find why the login test fails and fix it
/review
/artifacts last
/commit fix: correct login validation

You bring your own model API key. VS Code must remain open on the machine with your project.

🚀 Features

• Autonomous Agent Loop: Continuously plans and executes multi-step tasks. The agent reads your workspace, runs bash commands, edits files, and loops until the task is done.
• Remote Control Channels: Use Telegram and WhatsApp channels for remote task execution from your phone.
• Multi-Provider AI: Works with OpenRouter, OpenAI, Anthropic Claude, Google Gemini, MiniMax, Moonshot (Kimi), and self-hosted Ollama models.
• Structured Logs UI:
  • Grouped log view with typed rows (status, tool:start, tool:done, channel, llm, etc.)
  • Tool invocations are visible in the task timeline, including remote channel runs
  • Status verbosity modes: minimal, normal, debug
  • Filter chips and query filter for fast navigation in long runs
  • Collapse/expand all grouped nodes
• Native Coding Toolchain:
  • read_file / write_file / edit_file (exact-line replace)
  • fetch_url (read readable text from any HTTP website without external APIs)
  • diagnostics (instantly read VS Code TypeScript/ESLint errors without running compilers)
  • glob for file discovery
  • grep (ripgrep with Node.js fallback)
  • bash with captured stdout/stderr and a timeout
  • workspace context: active file, cursor, open documents
• Configurable Response Style: concise, normal, detailed.
• Session History: Recent conversation history is saved per workspace.
• Connection Check: Test the selected model and API key before running a task; the check sends no project files and runs no tools.
• Model Autocomplete: Search current model choices as you type, use the keyboard to select, and filter OpenRouter to free models.
• Credential Storage: API keys and Telegram tokens are kept in VS Code SecretStorage.

💡 Why a VS Code Extension?

1. Zero Context Switching: TeleCode AI lives directly in your sidebar. No terminal juggling.
2. Native Context: The agent instantly knows your Active File, Cursor Position, and Selected Text.
3. Instant Feedback: The agent modifies the exact files you have open. See live diffs without jumping between apps.
4. Remote Control via Telegram:
   • 🚑 Emergency fix on the go: A bug is reported. Message the bot from your phone, inspect its result, then request a commit.
   • 🛏 Work from bed: Your dev environment is on your office PC. Just leave VS Code open and instruct remotely.
   • 🚶 Background tasks: Ask the agent to build a feature, go for a walk. Code is waiting when you return.

🛠 Getting Started

1. Install from the VS Code Marketplace and open a project folder.
2. Open TeleCode AI from the Activity Bar, then Settings → Inference.
3. Choose a provider and enter its API key. Click the model field to load choices, then type to filter; use ↑/↓ and Enter to select. OpenRouter also has a Free models only filter. You can enter any model ID manually or use ↻ to refresh the list. The default is OpenRouter's openrouter/free; free models have provider rate limits and varying availability.
4. Click Test connection. This makes one short model request, which may incur provider usage charges.
5. Save settings, start the agent, and send a small task such as Explain the project structure without changing files.

Requires VS Code 1.107+. Other compatible IDEs need an extension host running Node.js 22.19+. Ollama uses your local server and does not require an API key.

Telegram

Create a bot with @BotFather. In TeleCode settings, enable Telegram and enter the bot token and your numeric Chat ID, then save. Commands remain blocked until a valid allowed chat is configured. Send /help to the bot to see the commands.

WhatsApp

Enable WhatsApp in TeleCode settings and save. Open Logs and scan the QR code with WhatsApp → Linked Devices → Link a Device. The default access mode accepts commands only from your own linked account. Keep VS Code open for remote tasks and schedules.

Permissions and Git changes

• Strict provides read/search tools. Balanced adds file editing and shell commands. Power also allows tool paths outside the project.
• Shell commands are not sandboxed. A working-directory restriction does not prevent a command from accessing other paths.
• Tool execution is autonomous; interactive tool approvals are not supported. The legacy autoApprove=false setting prevents agent startup.
• maxSteps limits actual tool calls per task. Provider failures and interrupted runs are reported as failures or interruptions.
• Automatic task commit/revert requires a recorded clean starting working tree. It also refuses files changed after the task finished. For existing uncommitted work, review and manage the changes manually in Git.

📱 Remote Commands

TeleCode AI is designed to be useful from messengers first. Telegram and WhatsApp share the same practical remote flows:

• run/review/checks for daily task execution
• queue/history/task/cancel for task control
• commit/revert/artifacts for result handoff
• memory/remember/forget for workspace-scoped project memory
• git/changes/diff/logs for lightweight remote inspection
• schedule for periodic remote tasks

Typical examples:

/run fix failing tests in taskRunner and show me the result
/review
/history 10 failed
/task last
/artifacts last
/memory
/remember repo uses pnpm only
/git status
/schedule every 1 send me the latest git commit title

The VS Code panel remains useful for logs and local runs, but the main operational loop can be fully driven from Telegram or WhatsApp.

⚙️ Development

npm install       # Install dependencies
npm run build:all # Build extension and webview
npm test          # Run test suite
npm run test:coverage # Run tests with coverage report
npm run watch     # Watch mode

For hot reload: open Run and Debug → start Run Extension (Watch).

⚙️ Settings

⏱ Remote Scheduler

You can create recurring remote tasks directly from Telegram or WhatsApp:

/schedule every 1 send me the current date and time
/schedule every 5 send me the latest git commit title
/schedule
/schedule run 1
/schedule pause 1
/schedule resume 1
/schedule remove 1

Notes:

• Minimum interval is 1 minute.
• Schedules are persisted in the workspace under .telecode/remote-schedules.json.
• Scheduled runs reuse the same task queue, so they do not bypass active manual tasks.

🧱 Architecture

TeleCode uses @earendil-works/pi-agent-core and @earendil-works/pi-ai. src/extension.ts wires the VS Code UI, lifecycle and channel controllers. Tools live in src/tools/definitions/; Telegram and WhatsApp use shared runtime and task queue helpers.

📄 License

MIT © Uginy