---
title: Vision
description: The vision and philosophy behind bobbin as a local-first code context engine
tags: [appendix, vision]
status: draft
category: appendix
related: [appendix/roadmap.md, architecture/overview.md]
---
# Bobbin Vision & Mission
## Mission Statement
> **search · coupling · context — local · private · fast**
Bobbin is a local-first context engine that gives developers and AI agents deep, structured access to codebases without sending data to the cloud.
## Vision
In the era of AI-assisted development, context is everything. Current tools treat code as flat text, losing the rich structural and temporal information that makes codebases understandable. Bobbin changes this by treating code as a living, evolving artifact with history, structure, and relationships.
**Bobbin enables "Temporal RAG"** - retrieval that understands not just what code says, but how it evolved and what changes together.
## Core Principles
### 1. Local-First, Always
- All indexing and search happens on your machine
- No data leaves your environment
- Works offline, works air-gapped
- Your code stays yours
### 2. Structure-Aware
- Code is parsed, not chunked arbitrarily
- Functions, classes, and modules are first-class citizens
- Respects language semantics via Tree-sitter
### 3. Temporally-Aware
- Git history is a retrieval signal, not just version control
- Files that change together are semantically linked
- Understand "what usually changes when X changes"
### 4. Agent-Ready, Human-Friendly
- CLI interface works for both humans and AI agents
- No special agent protocols required (MCP optional)
- Simple, composable commands
## What Bobbin Is
- A **code indexer** that understands syntax structure
- A **documentation indexer** for markdown/text files
- A **git forensics engine** that tracks file relationships over time
- A **semantic search** engine using local embeddings
- A **keyword search** engine for precise lookups
- A **context aggregator** that pulls related information together
## What Bobbin Is Not
- Not a task manager or agent harness
- Not an IDE extension (headless by design)
- Not a cloud service
- Not an AI agent itself - it serves agents
## Key Differentiators
| Chunking | Fixed token windows | AST-based structural units |
| History | HEAD only | Full git timeline |
| Relationships | Vector similarity only | Temporal coupling + similarity |
| Privacy | Often cloud-based | Strictly local |
| Runtime | Client-server | Embedded/serverless |
## Target Users
1. **AI Coding Agents** - Claude Code, Cursor, Aider, custom agents
2. **Developers** - Direct CLI usage for code exploration
3. **Tool Builders** - Foundation for context-aware dev tools
4. **Agent Harnesses** - Middleware integration for orchestration tools
## Success Metrics
- Sub-second query latency on repos up to 1M LOC
- Zero network calls during operation
- Retrieval precision that matches or exceeds cloud alternatives
- Seamless adoption in existing workflows
## Technology Choices
| Language | Rust | Performance, memory safety, ecosystem (Tree-sitter, LanceDB) |
| Vector Store | LanceDB | Embedded, serverless, git-friendly storage |
| Parser | Tree-sitter | Incremental, multi-language, battle-tested |
| Embeddings | Local ONNX (all-MiniLM-L6-v2) | Fast CPU inference, no API dependency |
---
*Bobbin: search · coupling · context — local · private · fast.*