Aller au contenu
TakoAPI
Retour aux skills

edgehdf5-memory

par osobh

Search & Research
—Pas encore noté
Ta note

Résumé

HDF5-backed persistent cognitive memory for AI agents.

Description

Ce que fait ce skill

HDF5-backed persistent memory store for AI agents. Saves conversation chunks with embedding vectors and retrieves them via hybrid vector + BM25 search. Supports WAL writes, Hebbian learning, temporal decay, and AGENTS.md generation.

Stores embeddings directly in a standard HDF5 file, so memory is portable and readable by any HDF5 library without a separate database server.

Quand l'utiliser

  • Persisting agent conversation history across sessions
  • Retrieving relevant past exchanges by semantic similarity
  • Backing up and snapshotting agent memory files
  • Generating AGENTS.md summaries from stored memory
  • Exporting all memory entries for inspection or migration

Exemple de workflow

ENTRÉE

User asks: 'What did we discuss about the weather last week?'

AGENT
  1. 1

    Embed the query text using an external embedding model

  2. 2

    Run edgehdf5 search with the embedding vector and query text against the .h5 file

  3. 3

    Retrieve top-k matching chunks ranked by hybrid vector + BM25 score

  4. 4

    Inject returned conversation chunks into the prompt as context

  5. 5

    Generate a response grounded in the recalled memory

SORTIE

Agent responds with relevant details from the stored conversation history

Prérequis

  • edgehdf5 CLI binary installed via cargo install edgehdf5-cli
  • EDGEHDF5_PATH environment variable set to the target .h5 file path
  • External embedding model or API to generate vectors before saving or searching

Installation

clawhub install osobh/edgehdf5-memory

Exécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.

Liens