Coding Intelligence x XolvedAI

A VS Code coding agent powered by xAI Grok 4.20. Streams responses in place, calls tools on your workspace with per-action approval, and exposes xAI's server-side tools (web search, code interpreter, X search) alongside vision input.

Features

• Streaming chat — one assistant bubble that fills in place, markdown formatted.
• Client-side tools (with approval):
  • list_files, read_file, search_files — read-only, auto-approved
  • write_file, create_directory, apply_patch, run_terminal — destructive, prompt before running
• xAI server-side tools (toggles in the header):
  • 🌐 web_search — live web results
  • 🐍 code_interpreter — Python sandbox
  • 𝕏 x_search — X posts and threads
• Vision — paste or attach images (PNG, JPG, WebP, GIF); Grok sees them. Provider-side size limit is 20 MB per image.
• Reasoning effort — Fast / Balanced / Deep toggle.
• Efficient multi-turn — uses previous_response_id so follow-ups don't resend history.
• Workspace-scoped path safety — file tools reject paths that escape the workspace root.

Installation

1. Install the .vsix:

   code --install-extension coding-intelligence-x-xolvedai-1.15.1.vsix
   

2. Set your xAI API key (see below).
3. Press Ctrl+Shift+G (macOS: Cmd+Shift+G) to open the agent panel.

API key

The extension reads XAI_API_KEY from either the environment or a .env.local file at the workspace root. See .env.example.

# Option A — env var
export XAI_API_KEY=xai-...

# Option B — workspace file (gitignored)
cp .env.example .env.local
# edit .env.local, paste your key

Get a key at console.x.ai.

Project memory

If a XolvedAI-Coding.md file exists at the workspace root, its contents are appended to the system instructions every turn. Use it for project-specific context (architecture notes, coding conventions, domain language).

Approval behavior

• Read-only tools run without prompting.
• write_file, create_directory, apply_patch, run_terminal show an inline approval card in the chat. Click Approve or Deny; denied calls return user denied to the model so it can adapt.
• run_terminal has a 30s timeout and output capped at 16 KB.

Keyboard shortcut

Architecture

src/
  extension.ts    VS Code activation, webview lifecycle, message routing
  xaiClient.ts    POST /v1/responses, SSE parser, agent turn loop
  tools.ts        Tool schemas + guarded local executors
  approval.ts     Approval registry (bridges webview UI ↔ agent loop)
  webview.ts      Chat UI (HTML/CSS/JS, CSP-locked, no remote scripts)

The client talks to https://api.x.ai/v1/responses with stream: true and parses SSE events (response.output_text.delta, response.output_item.added for function calls, response.function_call_arguments.delta, response.completed). Client-side tool calls are looped back as function_call_output items until the model emits no more calls.

Development

npm ci
npm run compile        # tsc
npm run watch          # tsc --watch
npx vsce package       # produce .vsix

License

MIT. See LICENSE.txt.