Documentation

Welcome to Backbencher. Unlike traditional AI tools that search text, Backbencher uses AST (Abstract Syntax Trees) to understand your code structure.

Installation

Backbencher is distributed as a standalone binary. No Node.js or Python required.

macOS & Linux

curl -fsSL https://backbencher.cc/install.sh | bash

Windows (PowerShell)

irm https://backbencher.cc/install.ps1 | iex

Quickstart

1. Index your Brain (Required)
Navigate to your project directory and build the SQLite AST database.
cd your-project && bb brain index
2. Launch Backbencher
Start the interactive terminal UI.
bb
3. Connect Provider
Inside the TUI, type the slash command to configure your API keys (e.g. Anthropic, OpenAI).
/connect

Upgrading Backbencher

Backbencher includes a built-in command to safely upgrade your standalone binary and dependencies to the latest release.

Auto-Upgrade Command

Run this anywhere in your terminal to fetch the latest version securely.

bb upgrade

Use bb as an MCP Server

You can expose all of Backbencher's native AST intelligence (e.g., bb_select, bb_relationships) directly to your favorite GUI editor like Cursor or Claude Desktop using the Model Context Protocol (MCP).

Claude Desktop Configuration

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "backbencher": {
      "command": "bb",
      "args": ["mcp-server"]
    }
  }
}

License Required

MCP server access is available to all Backbencher users. No license required.

Powered by Opencode

Backbencher is built on top of the Opencode framework. This means that all industry-standard configuration for LLMs, MCPs (Model Context Protocol), and Skills work exactly the same way.

Free Models Provided by OpenCode

Thanks to OpenCode! The free models provided in this tool are from OpenCode, not from us. We just used their open-source code to create bb, that's it. All credit for the free model access goes to them.

LLM Configuration

Easily switch between Anthropic, OpenAI, or local models via `/connect` or config files.

MCP Servers

Native support for Model Context Protocol. Connect your existing MCP tools instantly.

View Opencode Documentation

Native Brain Tools

Autonomous Usage

You do not need to call these tools manually. When you type a prompt (e.g., "Fix the auth logic in the login function"), the LLM autonomously selects these AST tools to search, extract, and update your code.

However, for power users, these commands are available in the CLI and TUI for manual surgical execution.

`bb_search`

Semantically and syntactically searches the project's native brain index using AST intelligence.

Scope: project, functions, classes, semantic, exact_text
Usage: Finding symbols, codebase exploration, dependency discovery.

`bb_select`

Fetches a compiled Context Package for a specific symbol using surgical AST precision. It dynamically extracts isolated snippets directly from the file.

Usage: Replaces the primitive \`read\` tool. Used to understand exact implementations without wasting tokens on 2,000-line files.

`bb_update`

Semantically updates a function, class, or symbol by replacing it entirely based on exact Tree-sitter byte coordinates.

Safety: Ensures zero collateral damage to surrounding indentation or logic.
Usage: Rewriting discrete architectural symbols found via \`bb_search\`.

`bb_relationships`

Explores the dependency graph instantly using the SQLite brain.

Features: Finds Callers, Callees, Class inheritance, and Impact analysis.
Usage: "What will break if I change this function?"