EverMind-AI/EverOS

EverOS – Persistent, Markdown‑backed memory for AI agents

What it is – EverOS is a Python library that gives any LLM‑powered agent a local‑first long‑term memory layer. It stores conversations, files, and agent “episodes” as plain‑text Markdown files, then keeps a SQLite index and a LanceDB vector index in sync so the data can be searched quickly.

Why it matters – Most agent‑memory tools hide the data behind opaque databases or cloud services. EverOS makes the source of truth human‑readable, version‑controlled (Git‑friendly), and editable with any editor. This lets developers inspect, diff, and evolve memories without a separate dashboard.

Core ideas

Feature	How EverOS does it
Markdown as source of truth	Every episode, profile, or knowledge page is a .md file on disk. The file can be edited directly; a file‑watcher updates the SQLite/LanceDB indexes automatically.
Three‑part local stack	No external services are required: • Markdown files for persistence • SQLite for relational queries (e.g., by user, app, project) • LanceDB for fast vector search when an embedding provider is configured.
Separate user & agent tracks	episodes/profile store a user’s personal history, while cases/skills store an agent’s learned abilities.
Orthogonal retrieval	Queries can be scoped by user_id, agent_id, app_id, project_id, and session_id – giving fine‑grained control over what memory is consulted.
Knowledge Wiki	Editable Markdown knowledge pages with a taxonomy, CRUD API, and topic search are part of the same memory store.
Reflection / offline evolution	After a session ends, EverOS can merge related episodes, refine profiles, and extract reusable skills before the next run.

Quick start (one‑key setup)

1. Install – pip install everos (or uv pip install everos).
2. Run the demo – everos demo shows the ingest → index → recall loop without any API key.
3. Configure an LLM – everos init creates ~/.everos/everos.toml. Fill in an OpenRouter API key (e.g., model = "openai/gpt-4.1-mini").
4. Start the server – everos server start (exposes /api/v2/memory/*).
5. Add & retrieve memory – POST JSON to /api/v2/memory/add, /flush, and /search. With only the LLM config you get keyword search; add an [embedding] section to enable hybrid (vector + keyword) search, reflection, and skill extraction.

Optional extensions

Extension	What it adds
everos[multimodal]	Image, PDF, audio, and office‑document ingestion via everalgo-parser. Requires LibreOffice for .doc/.ppt/.xls conversion.
Embedding provider ([embedding])	Hybrid search, memory clustering, and automatic skill extraction.
Rerank provider ([rerank])	Agentic search and Knowledge‑Wiki‑style ranking.

Typical use cases (as shown in the README)

• Smart‑glass assistants – persistent context for sports or wearable agents.
• AI coding assistants – long‑term recall of code snippets, project history, and developer preferences.
• Data‑analysis technicians – agents that remember past explorations of high‑dimensional time‑series data.
• Personal memory aids – apps for Alzheimer’s support, tasting journals, or creative brainstorming that keep a living log of user experiences.
• Multi‑agent orchestration platforms – a shared memory backend that lets several agents (Claude, Gemini, OpenCode, etc.) collaborate via a common knowledge base.

Who should use it

• Developers building LLM agents who need a lightweight, offline‑first memory store that stays in version control.
• Product teams that want to expose a memory API to downstream apps (e.g., browsers, smart glasses, IDE plugins) without managing external vector databases.
• Researchers exploring memory‑driven prompting, reflection, or long‑horizon agent behavior.

Getting started in code (minimal example)

from everos import EverOSClient

client = EverOSClient(root="~/.everos")
# Add a short conversation
client.add_memory(
    session_id="demo-001",
    app_id="demo",
    project_id="demo",
    messages=[
        {"sender_id": "alice", "role": "user", "content": "I love climbing in Yosemite."},
        {"sender_id": "alice", "role": "assistant", "content": "Sounds fun!"},
    ],
)
client.flush_memory(session_id="demo-001")

# Keyword search
results = client.search_memory(
    user_id="alice",
    query="Where do I like to climb?",
    method="keyword",
    top_k=5,
)
print(results)

The call writes Markdown files under ~/.everos, updates the SQLite/LanceDB indexes, and returns the matching episode.

Ecosystem

EverOS ships with ready‑made integrations for:

• DeepSeek Harness, Hermes, OpenClaw, Raven, and Dify (via the EverMind-AI/plugins repo).
• Community projects listed in the README (e.g., AIUI Sports Agents, Hive Orchestrator, EverMem sync tool, etc.).

License & contribution

The repository follows a typical open‑source model (the README points to make test, uv sync, and a CONTRIBUTING guide). Contributions are welcomed via pull requests; the project maintains a Discord and WeChat community for support.

---

Bottom line – EverOS is a genuine, production‑ready library that lets you give any LLM‑based agent a durable, editable, and searchable memory store without relying on external cloud services. Its emphasis on Markdown‑backed persistence makes it especially attractive for developers who want transparency and version control over their agents’ long‑term knowledge.