Give your AI coding assistant persistent, semantic memory
Three-tier memory architecture solves context bloat for Claude Code and Copilot CLI users.
Persistent structural memory for AI coding agents. Turns your repo into a fast, MCP-native knowledge graph so assistants stop grepping and start querying.
Persistent graph memory via MCP beats grep-and-trace for 5000+ file repos.
Developers working with large codebases or monorepos
Sourcegraph · Cursor · Continue
Three-tier memory architecture solves context bloat for Claude Code and Copilot CLI users.
MCP-native persistent memory prevents re-explaining projects to Claude every session.
Persistent memory for Cursor and Claude when every chat resets to zero.
Handy CLI that actually pulls a repo and runs a battery of concrete heuristics — repo age, README keywords/emoji counts, commit velocity, presence of LLM-tool files, and simple commit-statistics — to guess whether code was prompt-expanded. The rule set is intentionally arbitrary and extensible, which makes this a useful starting point for audits or CI checks, but expect false positives and plan to tune or add rules for your context.
Local-first MCP server with 26ms P95 search when cloud-based code intel requires uploads.
49 MCP tools and 94% token reduction, but 'chat with codebase' is a crowded category.