ChristopherKahler/base

What it solves

AI coding agents often suffer from "amnesia," treating every session as a fresh start without knowledge of your specific codebase, project history, or architectural decisions. This leads to agents relying on generic guesses or repetitive file searching (grepping) rather than understanding the context of your work.

How it works

It functions as a local, persistent knowledge graph engine that integrates directly into your agent's workflow (specifically Claude Code). It maps your workspace into a graph using tree-sitter for code structure and NQuads for business logic, decisions, and documentation. When you interact with an agent, the tool automatically injects the most relevant "slice" of this graph—such as domain rules, file dependencies, or recent project decisions—directly into the agent's context window at critical moments, such as when a session starts or before a tool is executed.

Who it’s for

Developers who use AI coding agents and want them to have a persistent, accurate, and context-aware memory of their specific projects, coding standards, and architectural rationale without needing to manually paste documentation or prompt the agent repeatedly.

Highlights

• Proactive Context Injection: Automatically feeds relevant information to the agent based on the current task, rather than waiting for a user query.
• Multi-Layered Memory: Tracks code structure (via AST), project milestones, business decisions, and documentation in a single, queryable graph.
• Persistent Sessions: Supports handoffs and forks, allowing agents to resume work with full context across different sessions.
• Local-First Architecture: Operates as a single Rust binary that keeps all data local to your machine, using git-friendly NQuads files for persistence.
• Language Agnostic: Supports over 35 programming languages via tree-sitter.