COGNITIVEX · THE DEVELOPER PLATFORM

Build on the Large Cognition Model

The AI memory SDK and cognitive infrastructure for agents and apps. One install gives any LLM living memory (recall, learning, dreaming, and reflection) through the cogx SDK, an MCP server, or a plain HTTP API.

WHAT YOU'RE BUILDING ON

Cognitive infrastructure, not another wrapper.

An LLM does query → model → response → forget. Every call starts from zero. The Large Cognition Model closes the loop: query → living memory → reasoning → learning → evolution. Your app keeps what it learns and gets sharper with use.

CognitiveX is the platform that runs that loop for you. You don't train a model, run a vector database, or stitch together a retrieval pipeline. You write memories in, ask questions out, and the cognitive layer handles embedding, ranking, consolidation, and recall across every session, for one agent or a whole fleet of them.

FIVE MINUTES TO LIVING MEMORY

Install

Three surfaces, one backend at api.cognitivx.io. Pick whichever fits how you build. They read and write the same memory.

1 · Wire it into Claude (or any MCP client)

terminal

# install the CLI
npm i -g @cognitivx/cli

# register the cogx MCP server with Claude
cogx mcp install claude

Claude (or Cursor, or any MCP-capable agent) now has remember, recall, reflect, and dream as native tools. Nothing else to host.

2 · Use the SDK in your own app

typescript · the real @cognitivx/sdk surface

import { configureApiClient, memory } from "@cognitivx/sdk";

configureApiClient({ apiKey: process.env.COGNITIVEX_API_KEY });

// store: free to write, no schema to define
await memory.remember({
  content: "User prefers terse answers and ships on Fridays.",
  memory_type: "semantic",
});

// recall: semantic search across every past session
const hits = await memory.recall({ query: "how does this user like to work?" });

3 · Or just call the HTTP API

bash · the real HTTP API

curl https://api.cognitivx.io/api/recall \
  -H "Authorization: Bearer $COGNITIVEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "query": "what did we decide about pricing?", "deep": true }'

Full reference, auth, and language guides live at docs.cognitivx.io.

WHAT YOU CAN BUILD

From a memory call to a system that learns.

Agents that remember

Give a coding or support agent recall of every past session. Decisions, fixes, and user preferences carry forward instead of resetting each conversation.
Share one memory across a fleet: what one agent learns, the others can recall through the same MCP.

Apps with a memory layer

Drop persistent, per-user memory into a product without standing up a vector DB, embedding pipeline, or retrieval ranking of your own.
Four memory tiers (semantic, episodic, procedural, foundational) so facts, events, how-tos, and identity are stored and recalled distinctly.

Systems that improve

Overnight dream consolidation compresses and links memories; salience and pattern detection surface what matters across sessions.
Reflection and introspection let an agent reason about its own memory and state, not just the current prompt.

Same memory, three doors → cogx SDK · MCP server · HTTP API · all backed by api.cognitivx.io

COGNITIVE MODES

The verbs of the platform.

Every surface exposes the same core operations. Writing memories is free; recall spends recall credits scaled by how deep you ask the system to think.

Mode	What it does	Cost
remember	Store a fact, event, how-to, or identity anchor into the right memory tier.	free to write
recall	Semantic search across every session. Choose depth: foundational, standard, or deep.	1 · 3 · 10 credits
learn	Record an outcome signal so future recall weights what actually worked.	free
reflect	Introspect over the memory graph: state, narrative, and self-assessment.	standard
dream	Consolidate: compress redundant memories and synthesize relationships between them.	batch

Recall depth → Foundational 1 · Standard 3 · Deep 10 credits. Storing is always free. See pricing for credit allotments per plan.

WHERE IT PLUGS IN

Built for the tools you already use.

Because cogx ships as an MCP server, any MCP-capable client gets living memory with one command. Claude Code, Cursor, and other agent runtimes connect the same way.

QUESTIONS

Before you build

Is this a vector database?

No. Embeddings and similarity search are one layer underneath. The platform adds the parts a raw vector store doesn't: memory tiering, salience, learning signals, dream consolidation, and reflection. You get a cognitive layer, not a bucket of vectors.

Which LLM does it use?

The LCM treats the LLM as infrastructure, the text-rendering step at the end of a structured pipeline. The intelligence lives in the memory architecture and routing. Swapping models changes output quality, not system behavior. Bring your own model in your app; recall and consolidation work the same.

What does it cost to start?

The free tier includes 100 recall credits per month and up to 1,000 memories, enough to wire an agent and feel the difference. Storing memories is always free; you only spend credits on recall. Paid plans and pay-as-you-go metering scale from there. See pricing.

How is this different from other memory tools?

Most memory libraries store and retrieve. The LCM adds the loop (learning, consolidation, and reflection) so the system evolves with use rather than just recalling. See the comparison for a head-to-head.

Plug in living memory.

The cognitive infrastructure that makes every LLM smarter. Start free, no card.

Start building →Try iCog →