Back to directory

pi-session-search

A plugin for the pi coding agent that indexes and semantically searches a developer's own past coding sessions, working immediately with zero-configuration keyword search and adding hybrid semantic search once an embedding provider is configured.

pi-session-search

pi-session-search indexes past pi coding-agent sessions so they can be browsed and searched later: keyword search through SQLite's FTS5 extension works immediately with no configuration, and when an embedding provider is set up, results combine cosine similarity and BM25 keyword ranking through reciprocal rank fusion for better retrieval. Sessions can be listed and filtered by project, date range, or archive status, and any past conversation can be read back in full.

Indexing runs on a background worker thread so it never blocks pi's interface, parses session JSONL files on startup, and tracks changes incrementally afterward, answering from the existing index while a fresh sync is still warming up. At the start of a new session, it can inject a short, hidden list of recent related sessions before the user's first message, and it supports OpenAI, Mistral, AWS Bedrock, and local Ollama as embedding providers, or it can be installed together with a companion package that adds persistent memory and local knowledge search.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.