EverOS

Active
GitHub Python Apache-2.0

Description

EverOS is a platform for building, evaluating, and integrating long-term memory for self-evolving agents, enabling AI agents to continuously accumulate experience and optimize themselves.

Key Features

  • Markdown as source of truth — all memory persists as .md files, editable with Obsidian/Git and version-controlled
  • Lightweight three-piece storage — Markdown (truth) + SQLite (state/queue) + LanceDB (vector+BM25+scalar), no heavy dependencies
  • Hybrid retrieval — BM25 + vector (HNSW/IVF-PQ) + scalar filtering in a single LanceDB query
  • Cascade index sync — edit .md → file watcher → entry-level diff → LanceDB sync, sub-second
  • Dual-track memory — user-track (Episodes/Profiles) + agent-track (Cases/Skills)
  • Multi-modal support — inline text + small images/audio, large media via S3/OSS references

Use Cases

💡 Build long-term memory systems for AI agents to accumulate experience and user preferences across sessions
💡 Store and retrieve structured memory as Markdown files in personal knowledge management
💡 Provide context-enhanced responses for chat assistants based on conversation history
💡 Build self-evolving agents that continuously optimize decisions through Cases/Skills memory
💡 Maintain independent long-term memory spaces per user in multi-user scenarios

Tags

memory agent python framework evaluation

Categories

🧩 Agent Memory

Quick Start

Install with uv pip install everos, run everos init to generate .env config, edit API keys, then everos server start. Add memory via /api/v1/memory/add and retrieve with search endpoints. Optional everos[multimodal] for image/audio/PDF support.
Visit GitHub Visit Website

Related Projects