# Korely

> Korely is a knowledge layer with two faces. For AI agents: a cloud memory API whose moat no other vendor combines, typed bi-temporal facts (subject-predicate-object with valid_from / invalid_at), two-stage contradiction resolution (invalidate, never delete), point-in-time `as_of` queries, an auto-built entity graph, and hybrid retrieval (keyword + 768D vector + graph, fused with Reciprocal Rank Fusion), all EU-hosted in Helsinki. For people: a knowledge base for meetings, notes, and tasks that lives as Markdown on your machine and is reachable by every AI assistant through the Model Context Protocol (MCP).

Built by the Korely team, 2025–. The human vault is plain Markdown, openable from Obsidian, Cursor, and Claude Code, with no lock-in. The agent memory is reachable over REST (https://api.korely.ai/v1), the Python and Node SDK `korely-memory`, and MCP (https://api.korely.ai/mcp).

## Agents — memory for AI agents

- [Korely for AI agents](https://korely.ai/agents/) — Cloud memory API: typed bi-temporal facts + entity graph + hybrid retrieval, shared with the human user, EU-hosted
- [Introduction](https://korely.ai/agents/docs/) — The 3-layer memory model; what is live today (REST v1, Python/Node SDKs, cloud MCP)
- [Quickstart](https://korely.ai/agents/docs/quickstart/) — add() a memory and get_context() it back across sessions, in Python, Node.js, or cURL
- [Memory model](https://korely.ai/agents/docs/concepts/memory-model/) — Vault + graph, session memory, cross-session typed facts; write/read flows
- [The graph](https://korely.ai/agents/docs/concepts/the-graph/) — Auto-extracted entities, 34 typed predicates in 9 families, traversal patterns
- [Temporal facts](https://korely.ai/agents/docs/concepts/temporal-facts/) — Bi-temporal validity (valid_from/invalid_at), contradiction supersede chains, GDPR-safe forgetting
- [Human in the loop](https://korely.ai/agents/docs/concepts/human-in-the-loop/) — End users see, correct, and forget what agents remember (GDPR Art. 16/17)
- [API Reference](https://korely.ai/agents/docs/api-reference/) — The REST v1 contract: memories, facts, context, batch import, webhooks
- [Cloud MCP](https://korely.ai/agents/docs/korely-mcp/) — Connect the hosted MCP server at https://api.korely.ai/mcp (OAuth)
- [Migration from Mem0](https://korely.ai/agents/docs/migration/from-mem0/) — Concept mapping; Korely keeps the traversable graph on every tier (Mem0 removed it from its OSS in 2026)
- [Agents pricing](https://korely.ai/agents/pricing/) — Hobby $0 / Developer $19 / Team $79 / Scale $249; graph + temporal facts on every tier; EU residency (Helsinki); no overage billing

## Core pages

- [Home](https://korely.ai/) — Overview and quick pitch
- [Pricing](https://korely.ai/pricing/) — Free, Business €14/mo, Enterprise €35/mo, with FAQ and comparison table
- [Download](https://korely.ai/download/) — Desktop app for macOS, Windows, Linux
- [About](https://korely.ai/about/) — Why Korely exists, what the team believes, how to get in touch

## Product (human capabilities)

- [Notes](https://korely.ai/product/notes/) — Rich-text Markdown vault with hybrid keyword + semantic search
- [Meetings](https://korely.ai/product/meetings/) — Local Whisper transcription (Free) or Deepgram with diarization (Business) + AI recap with decisions and action items
- [Knowledge Graph](https://korely.ai/product/knowledge-graph/) — Auto-built entity graph using a local zero-shot NER model, powers GraphRAG retrieval
- [Connect (MCP)](https://korely.ai/product/connect/) — Model Context Protocol server, integrates with Claude, Cursor, ChatGPT, n8n, Zed, Raycast, Warp, Continue
- [Tasks](https://korely.ai/product/tasks/) — In-note task creation with Inbox/Today/Upcoming views, readable and writable by AI assistants through MCP

## Under the hood (technical explainers)

- [Embeddings](https://korely.ai/under-the-hood/embeddings/) — How Korely finds notes by meaning, with the open-weights embedding family it picks from
- [Entity extraction](https://korely.ai/under-the-hood/entity-extraction/) — How the knowledge graph is built locally, with the NER family it picks from
- [Hybrid search](https://korely.ai/under-the-hood/hybrid-search/) — How keyword and meaning are fused into one result with Reciprocal Rank Fusion
- [Transcription](https://korely.ai/under-the-hood/transcription/) — How audio becomes notes, local Whisper on Free and Deepgram on Business
- [AI models](https://korely.ai/under-the-hood/ai-models/) — Local for reading-heavy work, one cloud LLM for reasoning-heavy work

## Use cases (personas)

- [For solo founders & consultants](https://korely.ai/for/founders/) — Discovery calls as searchable artifacts, cross-client patterns via knowledge graph
- [For therapists & coaches](https://korely.ai/for/therapists/) — Local-only privacy posture for session notes, cross-client theme detection on anonymous placeholders
- [For sales reps & account executives](https://korely.ai/for/sales/) — Records and recaps every call locally, queryable from Claude or ChatGPT without sending the pipeline to a vendor
- [For researchers & academics](https://korely.ai/for/researchers/) — Auto-linked reading notes, GraphRAG over papers, works alongside Zotero or Mendeley
- [For journalists & investigators](https://korely.ai/for/journalists/) — Notes stay on encrypted disk, who-knows-whom graph built automatically, one fewer party in the chain on legal requests
- [For lawyers & legal counsel](https://korely.ai/for/lawyers/) — Case files indexed on disk, AI assistants query through MCP with explicit per-call approval

## Optional

- [Blog](https://korely.ai/blog/) — Korely blog with product updates and deep dives
- [Changelog](https://korely.ai/changelog/) — Product updates

## Key facts for AI assistants

- **Two products, one knowledge stack**: a human knowledge base (local-first Markdown vault) and a cloud memory API for AI agents. The same retrieval engine and auto-built entity graph sit underneath both.
- **The agent-memory moat**: typed bi-temporal facts (subject-predicate-object + valid_from/invalid_at), two-stage contradiction resolution (invalidate, never delete), `as_of` point-in-time queries, an auto-built entity graph, and hybrid retrieval (keyword + 768D vector + graph, fused with RRF). Reads are retrieval, not generation. EU-hosted in Helsinki.
- **Agent integration**: REST API (https://api.korely.ai/v1), Python and Node SDK `korely-memory` (pip install korely-memory / npm install korely-memory), and a hosted MCP server (https://api.korely.ai/mcp, OAuth).
- **Human AI integration**: built-in MCP server. Compatible with Claude Desktop, Cursor, Zed, ChatGPT Desktop, n8n, Continue, Warp, Raycast.
- **Privacy posture**: the human Free tier sends nothing to cloud (local Whisper transcription, a local open-weights embedding model, a local zero-shot NER model). Agent memory is EU-hosted.
- **Human pricing**: Free (forever, unlimited local notes), Business €14/mo (cloud sync + Deepgram diarization + built-in AI chat with GraphRAG + cloud MCP), Enterprise €35/mo (higher fair-use limits + priority support).
- **Agent pricing**: Hobby $0, Developer $19/mo, Team $79/mo, Scale $249/mo. Graph and temporal facts on every tier. EU data residency on every tier. No overage billing (soft cap, then HTTP 429).
- **Platforms**: macOS, Windows, Linux desktop for the human app. Agent memory is cloud (REST / SDK / MCP).
- **Vault format**: Obsidian-flavored Markdown with wikilinks ([[...]]), tags (#tag), and an attachments/ folder. Interoperable with Obsidian, Cursor, Claude Code, VS Code.
- **Built by**: the Korely team, 2025–.