Local Agent IDE (Base Scaffold)

This repository is the base scaffold for a local, private, plan-and-execute coding agent IDE.

Monorepo layout

• apps/vscode-extension: VS Code UI + command surface for chat/plan interactions.
• apps/agent-service: Local runtime service for planning/execution/tool orchestration.
• packages/core: Shared types and contracts used by extension and service.

Planned workflow

1. User sends a task in IDE chat.
2. Agent creates a step-by-step plan.
3. Agent executes each step and updates status checkmarks.
4. Agent returns summary, logs, and file changes.

Build Progress

• Current step: Step 8 - Reliability pass (follow-up handling, tool selection, per-project memory indexing).
• Next step: Strengthen multi-attempt planning and persistent memory retrieval quality.

Quick start

1. Install dependencies:
   • npm install
2. Build all packages:
   • npm run build
3. Start local agent service (dev):
   • npm run dev:agent
4. Build extension in watch mode:
   • npm run dev:ext

Model configuration

• Default endpoint: http://localhost:1234/v1
• Default model id: openai/gpt-oss-20b
• Optional env vars before running dev:agent:
  • MODEL_ENDPOINT
  • MODEL_ID
  • MODEL_API_KEY
  • APPROVAL_MODE (ask | auto | unrestricted)
  • BROWSER_HEADLESS (true | false, default true)
  • BROWSER_SLOW_MO_MS (number, default 0)
  • BROWSER_KEEP_OPEN_MS (number, default 0)

Notes

• This is a base structure only. Feature implementation will be added step by step.
• History persistence is now enabled in ./.local-agent-ide/history.json (chats + run snapshots).
• Browser automation tools are implemented with Playwright (browser_open, browser_click, browser_type, browser_wait_for, browser_screenshot, browser_eval).
• To watch browser actions live, run agent-service with BROWSER_HEADLESS=false (optionally add BROWSER_SLOW_MO_MS=200).
• If Chromium is not installed yet, run: npx playwright install chromium.