- Build memory from your tools: read Slack, GitHub, Linear, Notion, Google Drive, files and more into a knowledge graph.
- Give your agents memory: Claude Code, Codex, Cursor, ChatGPT and others recall from it and add what they learn.
- Ask it: search your memory in plain language from the app, the HTTP API or the Cloud SDK.
- Share it safely: decide which teams, people and agents can reach which datasets.
cognee.start_ui(). See Hosted vs. local.
Core concepts
What’s new at launch
Permissions & memory promotion
One place to see and set which teams, people, and agents can reach which datasets. Give or remove access in one plain sentence; nothing changes before you confirm. An agent never reaches more than its owner can.
Auto-ontology (Premium)
Replace the generic entity types (Person, Location, Document) with a taxonomy built from your own data, then refine it in plain text in the ontology chat editor. $99 / month.
Memory Review
Keep memory correct: settle conflicting facts, close gaps where memory could not answer well, and check upcoming events, all from the dashboard.
Rewards
Earn points for the work that builds and uses your memory. Levels unlock workspace credit.
Integrations
Read Slack, GitHub, Linear, Notion, Jira, Confluence, HubSpot, Microsoft Teams, and more into memory, and connect Claude Code, Codex, Cursor, ChatGPT, and other agents.
Quick start
Sign up and let onboarding build your first memory and connect an agent.
Memory promotion lifts key pieces from agent sessions and team memory from agent to person, team, and organisation, into a separate Promoted Memory dataset owned by the team that receives them. Promotion is reviewed rather than silent: admins decide each one under Needs attention on the Permissions page. Answering access requests on the page is rolling out.
Pricing
Ingestion typically costs about 2.5 times the tokens of the input, because extracting and deduplicating entities runs LLM calls over your text. See Account & Billing for auto recharge and the full breakdown.
Your workspace runs extraction, embeddings and answers on models Cognee manages: there is no LLM or embedding API key to bring, and the models can’t be changed in Cloud. See Do I need my own LLM key?.
Hosted Cloud vs. free local UI
- Hosted Cognee Cloud: managed compute and storage: a dedicated managed PostgreSQL database per workspace holds your data, its vector embeddings (pgvector) and the knowledge graph. See Architecture. Sign up at platform.cognee.ai.
- Local UI (free): run the UI on your own machine with
pip install cogneeand one function call.
cognee.serve() points it at your Cloud workspace.