Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

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

FeatureTraditional RAGBobbin
ChunkingFixed token windowsAST-based structural units
HistoryHEAD onlyFull git timeline
RelationshipsVector similarity onlyTemporal coupling + similarity
PrivacyOften cloud-basedStrictly local
RuntimeClient-serverEmbedded/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

ComponentChoiceRationale
LanguageRustPerformance, memory safety, ecosystem (Tree-sitter, LanceDB)
Vector StoreLanceDBEmbedded, serverless, git-friendly storage
ParserTree-sitterIncremental, multi-language, battle-tested
EmbeddingsLocal ONNX (all-MiniLM-L6-v2)Fast CPU inference, no API dependency

Bobbin: search · coupling · context — local · private · fast.