Introduction
Welcome to the CognitiveX documentation.
CognitiveX is the continuity layer for persistent memory and cognition. The core product, iCog, gives people and agents one durable memory across sessions, tools, and models.
The short version: iCog is not a model and not just a chatbot. It is a memory-first cognitive substrate. Models can change; identity, recall, provenance, and learned context remain attached to the user.
Product surfaces
| Surface | Host | Purpose |
|---|---|---|
| iCog Chat | icog.app | Primary memory-aware chat experience |
| API | api.cognitivx.io | FastAPI backend for apps, SDKs, and MCP |
| Console | console.cognitivx.io | Account, usage, billing, agents, and advanced controls |
| Developers | developers.cognitivx.io | API keys, playgrounds, logs, and integration tools |
| Docs | docs.cognitivx.io | API, SDK, MCP, and implementation guides |
| Forum | forum.cognitivx.io | Community support, announcements, feedback, and showcases |
| Research | research.cognitivx.io | Analysis, technical notes, and memory-system research |
| Auth | auth.cognitivx.io | Global SSO, OAuth, MFA, and cross-domain handoff |
What this documentation covers
- API — generated reference for every HTTP endpoint on
api.cognitivx.io. - SDK — TypeDoc-backed reference for
@cognitivx/sdk. - MCP — generated catalog of iCog tools exposed to MCP-capable clients.
- Concepts — how memory, cognition, identity continuity, and reflexes work.
- Guides — practical integration paths for chat apps, agents, MCP clients,
and the
cogxCLI.
Where to start
- If you're building an integration, jump to the Quickstart.
- If you're connecting Claude Desktop or Cursor, see MCP.
- If you need the backend shape, read the System map.
- If you are wiring auth, read Auth and identity.
- If you are using the terminal or a coding agent, use the cogx CLI guide.
- If you want to understand what iCog is doing under the hood, start with Cognition.
Operating principles
- Keep identity separate from the underlying LLM.
- Store durable facts as memory instead of relying on prompt history.
- Prefer literal recall with provenance when answering factual questions.
- Use absolute dates and concrete IDs when debugging integrations.
- Treat memory hygiene as part of application design: update stale facts, forget bad records, and preserve useful decisions.