SuperSmall
Context that keeps every agent on track.
SuperSmall is a local code-intelligence server, written in Rust, that serves focused context to any AI agent over the Model Context Protocol. It grounds that context in AST analysis of your code, the Story Crawler’s narrative of your codebase and history, and the Reference Library of coding and architecture patterns.
Small, focused context instead of your whole repo
Everything runs locally on your machine.
Served over MCP
One context engine for any AI agent that speaks the Model Context Protocol. Serve it locally with kandi-supersmall mcp.
AST symbol indexing
Tree-sitter parses your code into real symbols across 27 languages, so context comes from code structure, not text search.
Local embeddings
Semantic search runs on local ONNX embeddings. Your code never leaves your machine.
Story Crawler
Walks your codebase and its history to build narrative context agents can follow.
Reference Library
Coding and architecture patterns that keep agents on track.
Log mining
Reads your project’s own log streams, with redaction, to answer what actually failed at runtime.
Impact analysis
See what depends on a symbol before an agent changes it.
Persistent insights
Findings saved in one agent session are there for the next one.
Works with
Install
SuperSmall installs through the Kandi CLI from the KandiForge registry. It ships the native kandi-supersmall binary for macOS, Windows and Linux.