hermes-labs-ai/fidelis

Overview

Zero-LLM agent memory for Claude Code and AI agents, providing local-first BM25, dense-vector, and reciprocal-rank-fusion retrieval. Returns original passages verbatim by default. Available on PyPI as fidelis-memory under the MIT license.

Best for

Best for
Developers building local-first memory layers for AI agents in Python.

Use cases

• Add persistent memory to Claude Code sessions
• Retrieve exact source passages for AI agents
• Build local-first RAG without LLM dependencies

How to use

Install

uvx --from "fidelis-memory==0.2.0" fidelis mcp serve

Tools exposed

• fidelis_recall

Tested with

Codex, Claude Code, GitHub Copilot CLI, Gemini CLI, OpenClaw

Notes

Zero-LLM agent memory for Claude Code and AI agents, providing local-first BM25, dense-vector, and reciprocal-rank-fusion retrieval. Returns original passages verbatim by default. Available on PyPI as fidelis-memory under the MIT license.

23 stars on GitHub. Last updated 2026-09-21. Licensed MIT.

Use cases

• Add persistent memory to Claude Code sessions
• Retrieve exact source passages for AI agents
• Build local-first RAG without LLM dependencies

Pros

• Returns verbatim passages, preserving source fidelity
• Multiple retrieval strategies including reciprocal-rank fusion
• MIT-licensed and installable via pip

Cons

• Early-stage project with few GitHub stars
• Python-only, limiting integration to Python ecosystems
• No hosted option; requires self-managed local infrastructure

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.