XibeCode

AI-powered autonomous coding assistant for your terminal and browser.

Overview

XibeCode is a CLI agent that can read and edit code, run commands, and iterate on tasks from your terminal using LLMs. It includes a WebUI for a browser-based experience, AI-powered test generation, and multi-model support for both Anthropic and OpenAI-compatible providers.

What's new in v1.3.7

• Same features as v1.3.6 below; 1.3.7 is the current npm CLI (v1.3.6 was unpublished and cannot be republished per npm policy).

What's new in v1.3.6

• npm update hint: when you start xibecode / xibecode chat, the CLI may print a short stderr message if a newer version is published on npm (cached check). Set XIBECODE_DISABLE_UPDATE_CHECK=1 to turn it off.
• xibecode whats-new (changelog): compare your install to npm’s latest and open the package page for context.
• Docs: fuller reference for interactive slash commands (/setup, /config, cloud shortcuts, etc.) on the project docs site.

What's New in v0.9.1

Reliability and anti-hallucination upgrades

• Evidence-aware completion gate: the agent now requires grounded signals before declaring a task complete.
• Post-edit verification: write/edit operations are validated with read-back checks to reduce silent bad edits.
• Improved loop resistance: canonicalized loop detection blocks repeated or low-variation command/tool loops.
• Safer memory recall: recalled memories are labeled as unverified hints and filtered by minimum relevance score.
• Grounded context compaction: old context now keeps a Grounded Facts Ledger + conversation summary instead of blind truncation.

Better chat UX and tool transparency

• Large transcript retention in terminal chat (much less history loss during long sessions).
• Token-aware auto compaction around large context budgets (near 120k token estimate) instead of fixed message-count trimming.
• Richer tool rendering in chat: tool call arguments and result summaries are shown for better observability.
• Improved Markdown rendering for assistant output in the TUI.

Model and provider improvements

• /model picker refresh with provider/base-url-aware model loading.
• /format command to switch request wire format (auto, anthropic, openai) and persist it.
• Continued support for Anthropic, OpenAI-compatible providers, Gemini, and custom endpoints.

CI and release workflow hardening

• GitHub Actions reliability fixes for TypeScript + dependency install paths.
• No bundled Playwright — browser automation is via optional agent-browser / MCP; npm install does not pull browser binaries.
• Type declarations for marked-terminal included in source to prevent clean-CI type failures.

Installation

CLI (required)

npm install -g xibecode

From source:

git clone https://github.com/iotserver24/xibecode
cd xibecode
pnpm install
pnpm run build
npm link

New releases

• Run xibecode whats-new (alias xibecode changelog) to compare your installed version to the latest on npm.
• When you start xibecode or xibecode chat, the CLI may print a short notice if a newer version is on npm (check is cached; set XIBECODE_DISABLE_UPDATE_CHECK=1 to turn it off). Release notes for users are whatever you put on the npm package page (readme); you do not need GitHub releases for that.

Requirements

• Node.js 18+
• API key from Anthropic or OpenAI

Platforms: CLI and WebUI run on Linux (x64, ARM64), macOS (Intel, Apple Silicon), and Windows. For servers, Docker, or Raspberry Pi, use headless runs: xibecode run / xibecode run-pr with env-based config (no TUI). See DOCS.md for the full device matrix.

Quick Start

# Configure once
xibecode config --set-key YOUR_API_KEY

# Interactive terminal mode
xibecode chat

# Open WebUI in browser (recommended for beginners)
xibecode ui --open

# Autonomous run
xibecode run "Create an Express API with auth"

Main Commands

xibecode ui

NEW - Start the WebUI in your browser.

xibecode ui              # Start on localhost:3847
xibecode ui --open       # Auto-open browser
xibecode ui -p 8080      # Custom port

Features:

• v0.dev-style Layout - Activity bar (left) → Chat panel (resizable) → Code editor (right)
• Monaco Code Editor - Professional code editor with syntax highlighting and IntelliSense
• File Tree Explorer - Browse and open project files with recursive directory tree
• Multi-Terminal Tabs - Create/manage multiple shell sessions with + and X buttons
• Real PTY Terminal - Fully interactive bash/zsh with colors, tab-completion, vim/nano support
• Git Integration - Commit history graph, stage/unstage files, view diffs, write commits
• Settings Modal - Configure AI provider, display preferences, dev tools, and MCP servers
• MCP JSON Editor - Edit mcp-servers.json directly with Monaco syntax highlighting
• Custom Models - Add any AI model (Claude, GPT-4, DeepSeek, Llama) via dropdown + text input
• Real-time Chat - Streaming AI responses with markdown rendering
• Status Bar - Connection status, current mode, active AI model, cursor position
• Resizable Panels - Drag the divider between chat and code areas to adjust layout
• Slash Commands - Type / for commands and mode switching, @ for file references
• New Chat Button - Clear conversation with + button in chat input

xibecode run

Autonomous coding workflow.

xibecode run "Build a REST API with Express"
xibecode run "Fix the TypeScript errors" --verbose
xibecode run --file task.txt

Options:

• -f, --file <path> prompt from file
• -m, --model <model> model override
• --mode <mode> initial agent mode
• -b, --base-url <url> custom API URL
• -k, --api-key <key> API key override
• --provider <provider> anthropic or openai
• -d, --max-iterations <number> default 150 (0 = unlimited)
• -v, --verbose
• --cost-mode <mode> normal or economy (use cheaper model and lower iteration caps to save API cost)
• --dry-run
• --changed-only

xibecode run-pr

Autonomous coding with automatic branch + GitHub PR creation. The command:

1. Runs the full agent task (same as run)
2. Executes your project's test suite for verification
3. Creates a new branch (xibecode/<slug>-<timestamp> by default)
4. Commits + pushes the branch to origin
5. Opens a PR against the remote default branch via the GitHub CLI (gh)
6. Prints the PR URL and exits

The PR description now includes task summary, per-file changes rationale (diff-based), run stats (iterations/tool calls and token/cost when available), and verification results (test command + pass/fail + duration, including any self-correction retries). In --cost-mode economy, explanation is budgeted and may fall back to file + +/- counts.

Prerequisites:

• gh (GitHub CLI) must be installed and authenticated:

  gh auth login

Usage:

xibecode run-pr "Fix the TypeScript errors in src/core/agent.ts"
xibecode run-pr "Add input validation" --verbose
xibecode run-pr "Refactor utils" --branch feat/refactor-utils --draft
xibecode run-pr --file task.txt --skip-tests

Options:

• -f, --file <path> prompt from file
• -m, --model <model> model override
• -b, --base-url <url> custom API URL
• -k, --api-key <key> API key override
• --provider <provider> anthropic or openai
• -d, --max-iterations <number> default 150 (0 = unlimited)
• -v, --verbose
• --cost-mode <mode> normal or economy (save API cost)
• --branch <name> override generated branch name
• --title <title> override PR title
• --draft open PR as draft
• --skip-tests skip test verification before creating PR

Example output:

  ✅ Pull Request created successfully!

  PR URL: https://github.com/your-org/your-repo/pull/42

xibecode chat

Interactive terminal chat + tool use.

Options:

• -m, --model <model>
• -b, --base-url <url>
• -k, --api-key <key>
• --provider <provider>
• --cost-mode <mode> normal or economy
• --theme <theme>
• --session <id>

xibecode config

Manage saved config:

• --set-key, --set-url, --set-model, --set-provider
• --set-cost-mode <mode> set default cost mode: normal or economy
• --set-economy-model <model> model to use when cost mode is economy
• --show, --reset
• MCP helpers: --list-mcp-servers, --add-mcp-server, --remove-mcp-server

xibecode mcp

MCP server management:

• add, list, remove, file, edit, init, reload
• search, install, login (Smithery integration)

Core Features

• Autonomous multi-step agent loop - Completes complex tasks automatically
• Smart context gathering - Understands related files and imports
• Verified and line-based editing - Reliable code modifications
• Dry-run mode - Preview changes safely before applying
• Git-aware workflows - --changed-only, checkpoints, and reverts
• Test runner integration - Auto-detects Vitest, Jest, pytest, Go test
• MCP server integration - Extend capabilities with external tools
• Skill system - Built-in + custom markdown skills
• Session-aware chat - Persistent conversation history
• Themed terminal UI - Beautiful, customizable interface

WebUI

The WebUI provides a browser-based interface that syncs in real-time with the terminal.

# Start with both TUI and WebUI
xibecode chat

# WebUI opens automatically at http://localhost:3847

TUI-WebUI Sync

When you run xibecode chat, both interfaces are connected:

• Messages sent from TUI appear in WebUI (marked with "TUI")
• Messages sent from WebUI are processed by TUI
• Streaming responses show in both simultaneously
• Tool executions display in real-time

Slash Commands (/)

Type / in the input to open the command palette:

Commands:

| Command | Description | |---------|-------------| | /clear | Clear chat messages | | /help | Show available commands | | /diff | Show git diff | | /status | Show git status | | /test | Run project tests | | /format | Format code in project | | /reset | Reset chat session | | /files | List project files |

Modes:

| Mode | Icon | Description | |------|------|-------------| | /mode agent | 🤖 | Autonomous coding (default) | | /mode plan | 📝 | Interactive planning with web research | | /mode tester | 🧪 | Testing and QA | | /mode security | 🔒 | Security analysis | | /mode pentest | 🔓 | Penetration testing - run app and probe for vulnerabilities | | /mode review | 👀 | Code review | | /mode team_leader | 👑 | Coordinate team |

File References (@)

Type @ to browse and reference files:

• Shows project files and folders
• Filter by typing after @
• Select to include file path in message
• Helps AI understand which files to work with

Settings Panel

Click the ⚙️ Settings button to configure:

• Provider - Anthropic, OpenAI, or Custom
• Model - Select from available models
• Custom Model ID - For custom/local models
• API Key - Your provider API key
• Base URL - Custom API endpoint (for local LLMs)
• Session Info - Working directory, git branch

Features

• Markdown Rendering - Code blocks, bold, italic, lists, links
• Tool Execution - Shows each tool call with status (running/done/failed)
• Thinking Indicator - Spinner while AI is processing
• Responsive Design - Mobile-friendly layout and touch-friendly controls; works on phones, tablets, and desktop
• Real-time Streaming - See responses as they're generated

AI Test Generation

XibeCode can automatically generate comprehensive test suites for your code:

# Via CLI (in chat mode)
> generate tests for src/utils/helpers.ts

# Via WebUI
1. Go to "Test Generator" tab
2. Enter file path
3. Select framework (auto-detected)
4. Click "Generate Tests"

Features

• Multi-framework support - Vitest, Jest, Mocha, pytest, Go test
• Code analysis - Understands functions, classes, types
• Edge case generation - Null checks, empty strings, boundaries
• Mock setup - Automatic mock configuration
• Type checking tests - Verifies return types
• Error handling tests - Tests for exceptions

Example Output

import { describe, it, expect, vi } from 'vitest';
import { calculateTotal } from '../utils/helpers';

describe('calculateTotal', () => {
  it('should execute calculateTotal successfully', () => {
    expect(calculateTotal([])).toBeDefined();
  });

  it('should return correct type from calculateTotal', () => {
    expect(typeof calculateTotal([])).toBe('number');
  });

  it('should handle empty array', () => {
    expect(calculateTotal([])).toBe(0);
  });

  it('should handle errors in calculateTotal', () => {
    expect(() => calculateTotal(undefined)).toThrow();
  });
});

Browser testing (no bundled browser)

XibeCode does not ship Playwright or download Chromium. Legacy tool names (take_screenshot, preview_app, etc.) return a short message pointing you to run_command + agent-browser (install separately if you want it), your browser MCP, or fetch_url. For Playwright E2E, add @playwright/test to the target project and run it with run_command (for example pnpm exec playwright test).

Termux / Android: Upstream agent-browser does not publish an android-arm64 release binary, so a global npm i -g xibecode stays lean and skips that download. Install xibecode@latest (spell it latest, not latwst). For browser flows on Termux, rely on MCP, fetch_url, or tooling in your project.

Optional CLI (desktop Linux/macOS/Windows): npm i -g agent-browser when you need snapshot/click automation via run_command.

Configuration

Environment Variables

ANTHROPIC_API_KEY=sk-ant-...     # Anthropic API key
OPENAI_API_KEY=sk-...             # OpenAI API key
XIBECODE_MODEL=claude-sonnet-4-5-20250929  # Default model

Config File

Located at ~/.xibecode/config.json:

{
  "apiKey": "sk-ant-...",
  "model": "claude-sonnet-4-5-20250929",
  "provider": "anthropic",
  "maxIterations": 50,
  "theme": "default"
}

Available Models

| Model | Provider | Best For | |-------|----------|----------| | claude-sonnet-4-5-20250929 | Anthropic | General coding (default) | | claude-opus-4-5-20251101 | Anthropic | Complex reasoning | | claude-haiku-4-5-20251015 | Anthropic | Fast responses | | gpt-4o | OpenAI | General coding | | gpt-4o-mini | OpenAI | Fast responses | | o1-preview | OpenAI | Complex reasoning |

API

XibeCode provides a REST API when running the WebUI:

# Start the server
xibecode ui

# API endpoints
GET  /api/health          # Health check
GET  /api/config          # Get configuration
PUT  /api/config          # Update configuration
GET  /api/models          # List available models
GET  /api/project         # Get project info
GET  /api/git/status      # Get git status
GET  /api/git/diff        # Get git diff
POST /api/files/list      # List directory contents
POST /api/files/read      # Read file contents
POST /api/session/create  # Create chat session
POST /api/tests/generate  # Generate tests for file
POST /api/tests/analyze   # Analyze file for testable code
POST /api/tests/run       # Run project tests

Project Structure

xibecode/
├── src/
│   ├── core/           # Agent, tools, context
│   ├── commands/       # CLI commands
│   ├── utils/          # Config, git, safety
│   ├── tools/          # Test generator, browser
│   ├── webui/          # WebUI server
│   └── index.ts        # CLI entry point
├── site/               # Documentation site
└── tests/              # Test suites

Dependency Map

Project Docs

• CHANGELOG.md — release history
• FEATURES.md — feature deep dive
• PUBLISHING.md — npm release process

Support

• GitHub Issues
• Donate

License

Apache-2.0