The Dual-Engine
Codebase OS.
Stop relying on AI that blindly reads files and breaks your syntax. Backbencher uses surgical AST precision and a local SQLite Brain to navigate, understand, and modify your codebase with mathematical accuracy.
No credit card required. Preview ends in 30 days.
How the Brain Engine Works
When you first run bb in your project, something magical happens. Your entire codebase becomes a queryable, semantic graph.
JIT Brain Index
Backbencher parses every file through Tree-sitter AST parsers and compiles a local SQLite database containing every function, class, variable, and their relationships. No code leaves your machine.
[Brain] Indexing 2,385 files...
[Brain] 69,813 symbols mapped
[Ready] SQLite brain online
Surgical Context Extraction
Instead of reading entire files and wasting tokens, the AI queries the brain to extract only the exact byte-boundaries of the symbol it needs, along with its callers and callees.
// Extracts ONLY lines 142-167
// + 3 callers, 2 dependencies
Tokens saved: 94%
The AST Scalpel
When modifying code, Backbencher does not use primitive line-replacements. It surgically injects logic into the exact AST node boundaries, guaranteeing perfect syntax and indentation.
// Replaces bytes 4,521 - 4,892
// Syntax validated via Tree-sitter
Zero collateral damage
The Dual-Engine Architecture
Two engines working in perfect harmony. The Brain indexes. The AI reasons. The result is precision.
The SQLite Brain
A local, blazing-fast relational database storing every AST node, symbol location, and cross-file relationship in your codebase.
The AI Agent
Your chosen LLM (Claude, GPT, Ollama, etc.) equipped with native AST tools that query the Brain instead of blindly reading files.
Native Brain Tools
Forget grep,cat, andsed. These are the precision instruments the AI uses instead.
bb_searchSemantically search symbols, functions, and patterns across the entire codebase.
bb_selectExtract exact byte-bounded code snippets with caller/callee context.
bb_updateSurgically replace functions or classes using AST coordinates.
bb_relationshipsQuery the dependency graph: who calls what, and what calls whom.
bb_patternsDetect orphaned code, architectural violations, and complexity.
bb_securityScan for vulnerabilities, secrets, and taint flows across languages.
See It In Action
Watch how Backbencher handles a real modification request with surgical precision.
Get Started in Seconds
One command. No dependencies. The Brain builds itself on first run.
curl -fsSL https://backbencher.cc/install.sh | bashiwr https://backbencher.cc/install.ps1 -useb | iexFirst run will trigger bb_refresh to build your local SQLite brain. Takes around 10 seconds for a 50k LOC project.
Questions and Answers
What models can I use?
Bring Your Own Key (BYOK). Use Claude, GPT-4, Gemini, or even local models like Ollama and LMStudio. Backbencher is model-agnostic.
Is my code sent to the cloud?
Never. The SQLite Brain is built and stored 100% locally on your machine. Only your prompts are sent to your chosen LLM provider.
What languages are supported?
26+ languages including TypeScript, Python, Rust, Go, Java, C++, Ruby, PHP, Swift, Kotlin, and more via Tree-sitter parsers.
What happens after the preview?
You will have the option to subscribe for continued access. Preview users get priority pricing and lifetime benefits.
Ready to experience surgical precision?
Join the preview and see why developers are calling Backbencher the future of AI-assisted coding.