# thinqOS > thinqOS is the cognitive layer for AI: a Mind for every person and agent you work with. Each person and each agent gets a Mind: beliefs, preferences and lessons, each with a source. thinqOS decides which of them reach each task, under the right identity and authority, and the same Mind is reachable from more than one AI tool. thinqOS is in private preview. - MCP server: https://app.thinqos.com/mcp (streamable HTTP, bearer API key). The read tool is `recall_mind`. - CLI: `uv tool install thinqos` connects Claude Code, Codex, VS Code, Cline and Grok Build, with capture and recall hooks. - ChatGPT and Claude.ai: thinqOS connects to them as a custom MCP connection. Whether a plain chat calls it without being asked has not been measured. - Writes go through REST and are extracted asynchronously; a fact is recallable only after its extraction_status reaches "done". ## Connect - [MCP memory server: connect thinqOS to your agent.](https://thinqos.com/mcp-memory-server): thinqOS runs a real MCP server. What comes back is not everything you have stored: it is the part of your Mind that applies to this task, with sources. - [Portable AI memory across ChatGPT and Claude.](https://thinqos.com/ai-memory-across-chatgpt-and-claude): One Mind your identity owns, reachable from more than one AI tool, which decides what applies to the task instead of replaying everything you have said. - [Memory for Claude Code and Codex: one shared Mind](https://thinqos.com/memory-for-claude-code-and-codex): Your coding agents forget everything between sessions. thinqOS gives Claude Code and Codex one shared, persistent memory across tools and machines: auto-recalled before each answer, auto-captured after, with no plugin to babysit. - [Persistent memory for AI agents, over MCP and API](https://thinqos.com/for-builders): Keep your application. Connect a persistent Mind through thinqOS interfaces. - [Connect Claude Code, Codex or your app to thinqOS](https://thinqos.com/get-started): Four setup paths that do not overlap: get access, connect a coding assistant, start a business workflow, or run your own infrastructure. Pick the one you actually need. - [Run thinqOS on infrastructure you control.](https://thinqos.com/install): The self-hosting path is separate from connecting a coding assistant to the hosted service. ## Compare - [Mem0 alternative: thinqOS vs Mem0 compared](https://thinqos.com/compare/mem0): Mem0 is an open-source memory library with a hosted Platform. thinqOS decides which of your accumulated context reaches each task, under one identity. Where each is ahead, as of September 22, 2026. - [Zep alternative: thinqOS vs Zep and Graphiti compared](https://thinqos.com/compare/zep): Graphiti is an open-source temporal knowledge graph, and Zep Cloud is the hosted product built on it. thinqOS weighs beliefs and decides which of them reach each task, under one identity. Where each is ahead, as of September 22, 2026. - [Letta alternative: thinqOS vs Letta (MemGPT) compared](https://thinqos.com/compare/letta): Letta is an open-source agent harness whose agent edits its own memory. In thinqOS the runtime decides what reaches each task, not the model. Two different shapes, compared as of September 22, 2026. - [ChatGPT memory vs Claude and Gemini memory: thinqOS compared](https://thinqos.com/compare/chatgpt-claude-memory): ChatGPT, Claude and Gemini each remember you inside their own apps. thinqOS decides what reaches each task, and the same Mind is reachable from several tools. Where each is ahead, as of September 22, 2026. ## Product - [thinqOS platform: memory, cognition and agents](https://thinqos.com/platform): thinqOS connects what your AI knows to the work you need it to do. - [How thinqOS works: from memory to the next decision](https://thinqos.com/how-it-works): A Mind keeps what has been learned. Cognition decides how that experience applies now. - [Inspect what your AI agent knows and can do](https://thinqos.com/trust): Inspect memory, access and actions before entrusting an agent with important work. - [Add capacity without adding headcount.](https://thinqos.com/for-business): Give recurring work to AI agents that learn how your business works. Keep your team in charge of relationships and commitments. - [Start with the work you need done.](https://thinqos.com/use-cases): Choose a useful job, then decide what memory, cognition and tools it needs. ## Guides - [Memory recalls the price. Cognition works out the deal.](https://thinqos.com/theory/memory-and-cognition): Remembering what happened and deciding what to do are different jobs. - [What is a Mind?](https://thinqos.com/theory/what-is-a-mind): A persistent record of what someone believes, wants and has learned, with room for uncertainty and correction. - [When are notes enough for your AI?](https://thinqos.com/theory/notes-or-a-mind): A plain notebook can be the right answer. Know which job you need before adding another system. - [Your AI needs the right context.](https://thinqos.com/theory/context-and-cognition): A bigger prompt is not a substitute for deciding what belongs in it. - [Does giving AI more history improve its answers?](https://thinqos.com/theory/context-study): Our July 2026 replay compared five context strategies. The result supports careful selection, with important limits. - [How to compare AI memory systems.](https://thinqos.com/theory/memory-platforms): Test the behavior you need. Product labels and storage limits do not settle the decision. - [What should you expect from an AI agent?](https://thinqos.com/theory/agent-platforms): An agent needs more than a chat box. Ask how it works, remembers, acts and recovers. - [Carry project decisions between coding tools.](https://thinqos.com/theory/coding-tools): Make the earlier decision available, then verify that the next assistant uses it. - [How AI agents connect to your tools.](https://thinqos.com/theory/connect-tools): A connection gives an agent a way to act. Credentials, scope and verification determine what it should do. - [How AI agents work with other agents.](https://thinqos.com/theory/agents-working-together): A useful handoff has a job, a boundary and a result someone can check. - [Your agents should not share every secret.](https://thinqos.com/theory/separate-agents): Separate roles and context before several people rely on the same agent team. - [Who a fact is about is not who may see it.](https://thinqos.com/theory/private-and-shared): A useful privacy model separates subject, source, audience and operator. - [Your agent has context. You should too.](https://thinqos.com/theory/your-mind-and-your-agent): Your own preferences and an agent’s working knowledge serve different purposes. - [Does every agent need its own Mind?](https://thinqos.com/theory/how-many-minds): Create a specialist when it needs lasting judgment, not merely because another tool exists. - [AI help without another conversation.](https://thinqos.com/theory/help-without-another-chat): The best assistance may be a useful reminder at the moment you need it. - [What can AI learn between tasks?](https://thinqos.com/theory/learning-between-tasks): Reviewing experience is useful only if it improves later work. - [Learning is not permission to do anything.](https://thinqos.com/theory/learning-with-limits): Separate a new lesson from a change to an agent’s instructions, tools or authority. - [People, agents and organizations.](https://thinqos.com/theory/people-agents-organizations): Who owns the work, who performs it and who can access the information are separate questions. - [Change tools without starting from scratch.](https://thinqos.com/theory/open-connections): Portable memory and a portable running agent are different promises. - [What do we mean by cognition?](https://thinqos.com/theory/words-for-ai): Describe observable behavior before asking readers to adopt a new word. ## Optional - [Company](https://thinqos.com/company): who builds thinqOS. - [Brand](https://thinqos.com/brand): how to write the name: a lowercase "thinq" and an uppercase "OS".