# EverMind AI

> EverMind AI builds memory infrastructure for AI agents. Our core product, EverOS, is a Memory Operating System that transforms stateless LLMs into intelligent agents that remember, learn from experience, and evolve across sessions and platforms.

"Memory is the defining line between a tool and an agent."

Website: https://evermind.ai
Docs: https://docs.evermind.ai
Dashboard: https://everos.evermind.ai
GitHub: https://github.com/EverMind-AI/EverOS
Contact: contact@evermind.ai
Discord: https://discord.gg/gYep5nQRZJ
X: https://x.com/evermind
LinkedIn: https://www.linkedin.com/company/ai-evermind

## Products

### EverOS — Memory OS for AI Agents

The flagship product. Gives any LLM agent persistent, structured memory with automatic extraction, conflict resolution, and intelligent retrieval.

Key capabilities:
- **Structured memory extraction** — Automatically segments conversations into MemCells (episode + atomic facts + foresight + metadata) and clusters them into thematic MemScenes
- **Multimodal RAG (mRAG)** — Extracts memory from text, images, PDFs, audio, video, spreadsheets, and URLs
- **Self-evolving agent memory** — Records agent trajectories as Cases and distills patterns into reusable Skills
- **Multi-user attribution** — Tracks distinct profiles per participant in group conversations
- **Multi-agent shared memory** — Any agent authorized with the same user_id or group_id reads from and writes to the same memory namespace, with no explicit handoff required
- **Model-agnostic** — Memory is independent of any specific LLM; the same memory base is accessible regardless of which model the agent uses
- **Four retrieval methods** — Keyword (BM25), vector, hybrid (recommended default), and agentic (LLM-guided)
- **Memory Bank** — Visual interface for managing user, group, and agent memory

Deployment options:
- **EverOS Cloud** — Fully managed, production-ready. Start at https://everos.evermind.ai
- **EverOS Open Source** — Apache 2.0 licensed, self-host on your infrastructure. GitHub: https://github.com/EverMind-AI/EverOS

For integration details, API reference, and SDK quickstart:
- llms.txt (EverOS): https://docs.evermind.ai/llms.txt
- Full reference: https://docs.evermind.ai/llms-full.txt
- Docs: https://docs.evermind.ai

### EverMe — Personal Memory Hub

"Your Personal Memory Nebula." A consumer product exploring personal memory management for individuals. Currently in development.

## Technology Foundation

**EverCore** is the underlying research and systems layer powering EverOS. It includes:

- **EverMemOS** — Engram-inspired memory mechanism using MemCells and MemScenes. Implements a biological memory lifecycle: Episodic Trace Formation, Semantic Consolidation, and Reconstructive Recollection. Paper: https://arxiv.org/abs/2601.02163
- **HyperMem** — Hypergraph memory architecture for complex relational memory. Paper: https://arxiv.org/pdf/2604.08256
- **MSA (Memory Sparse Attention)** — End-to-end trainable architecture scaling to 100M+ tokens. Paper: https://evermind.ai/blogs/breaking-the-100m-token-limit-msa-architecture-achieves-efficient-end-to-end-long-term-memory-for-llms | Code: https://github.com/EverMind-AI/MSA

Both are research systems, not standalone products.

## Benchmarks (Memory Reasoning Accuracy)

Benchmarks reflect long-context memory reasoning and retrieval consistency, not generic embedding quality.

| Benchmark | Score | vs. Competitors |
|-----------|-------|----------------|
| LoCoMo | 93.05% | vs. Zep 85.22% |
| LongMemEval | 83.00% | vs. MemOS 77.80% |
| HaluMem | 93.04% | |
| HyperMem (LoCoMo) | 92.73% | |

## Use Cases

EverOS is designed for applications requiring long-term consistency and deep user understanding:
- AI companions and personal assistants
- Multi-agent systems coordination
- Customer support with cross-session history
- Educational and coaching agents
- Long-lived enterprise agents with persistent memory
- Wearable AI hardware integration

## Quick Integration (3 API Calls)

```
pip install everos

1. POST /api/v1/memories        — Add messages
2. POST /api/v1/memories/flush  — Trigger memory extraction
3. POST /api/v1/memories/search — Retrieve context (method="hybrid", top_k=5)
```

Full quickstart and SDK docs: https://docs.evermind.ai/cloud/quickstart

## Data and Privacy

Memory is namespaced by user_id and group_id — one user's memory is not accessible under a different scope. Users control all stored data; deletion available by memory_id or in bulk. Self-hosted deployment available via open source for full data locality.

## Links

- Product: https://evermind.ai/product
- Academy: https://evermind.ai/academy
- Ecosystem: https://evermind.ai/ecosystem
- About: https://evermind.ai/about-us
- Careers: https://evermind.ai/careers
- FAQ: https://evermind.ai/faqs
- Blog (SOTA benchmarks): https://evermind.ai/blogs/everos-hits-sota-performance-on-locomo
- Terms: https://evermind.ai/terms-of-service
- Privacy: https://evermind.ai/privacy-policy