The infrastructure behind
agents that actually know
Knowledge graphs, hybrid search, and a growing app ecosystem — everything your agents need to reason over what your organization actually knows.
TL;DR: Cortex provides eight core capabilities for AI agent memory: a knowledge graph with automated GraphRAG, conversation memory that persists across sessions, living sources that ingest the web and your Git repos, a Skills Hub built on the agentskills.io open standard, a growing app ecosystem, an API-first architecture with 60+ REST endpoints, enterprise-grade security, and natural-language Ask AI with source citations.
Knowledge Graph & GraphRAG
Your organization's memory, structured for agents
Cortex extracts entities, relationships, and concepts from your documents automatically. The result is a living knowledge graph that agents can traverse to reason across your entire organization — not just retrieve similar text.
- Automatic entity extraction (people, orgs, concepts, locations)
- Relationship discovery between entities
- Semantic entity resolution (merge duplicates)
- Multi-hop graph traversal for deep context retrieval
- Community detection and summarization
Key Metrics
Conversation Memory
The memory that survives the context window
Agents forget the moment the window closes. Cortex gives every conversation a durable memory — a curated context of rolling summaries, durable facts, open questions, and cited sources that carries from one session to the next, so your agent compounds what it learns instead of starting cold.
- Conversation memory that persists across sessions
- Context curator: rolling summaries, durable facts, open questions, and intent
- Stable source ledger — citations keep their identity across turns
- Memory fast-path answers follow-ups without re-running retrieval
- Backend stays stateless — memory travels with the client
Key Metrics
Living Sources
Connect the web and your repos as living knowledge
Your knowledge lives in more than uploads. Cortex harvests web pages into clean markdown and ingests entire Git repositories — code, docs, and wikis — keeping them in sync as they change. Connected repos are bidirectional: the agent can read them and act back by opening pull requests for human review.
- Web Import — harvest any URL into clean markdown, ingested like a document
- Git Integration for GitHub, GitLab & Gitea (incl. self-hosted)
- Bidirectional: ingest code, docs & wikis — and open pull requests
- Incremental, versioning-aware sync keeps the graph current
- YouTube channels and more via the app ecosystem
Key Metrics
Skills Hub
Extend retrieval — and take action — with the open standard
Cortex implements the agentskills.io open standard. Skills are lightweight instruction files that teach the retrieval engine new domain capabilities — and with the built-in http_request tool, they let your agent take real actions on external APIs, like creating tickets or opening PRs. Install a skill and every agent querying your Cortex gains that capability instantly.
- agentskills.io open standard — supported by 30+ agent products
- Built-in http_request tool — skills act on external APIs, not just retrieve
- Setup Wizard configures a skill's tokens and URLs in seconds — no .env edits
- Author custom skills in minutes with SKILL.md files
- Auto-activated, version-controlled, auditable, and portable
Key Metrics
App Ecosystem
A marketplace that deepens your graph
Browse a growing library of apps that plug into your knowledge graph. Source apps ingest knowledge from YouTube and more. Workflow apps unlock better interfaces and integrations. Every app you launch feeds the graph, and the graph gets smarter with every document.
- Source apps for YouTube and more
- Workflow apps for chat, dashboards, and integrations
- Premium apps for extra reach and capabilities — coming soon
- Request custom apps built for your team
- New apps added continuously
Key Metrics
API-First Architecture
Connect any agent, any framework, any model
60+ REST endpoints with full OpenAPI documentation. Integrate Cortex into any agent framework, build custom interfaces, or wire it into existing pipelines. When you switch stacks, the knowledge graph comes with you — full instance export and import included.
- 60+ REST API endpoints
- Full OpenAPI/Swagger documentation
- Streaming responses (SSE) for chat and deep research
- Background task tracking and per-key usage analytics
- Full instance export/import for portability
Key Metrics
Security & Isolation
Every customer on their own isolated instance, defended in depth
Security isn't a plan tier — it's how Cortex is built. Every customer runs on an isolated instance with its own knowledge graph and data store, never a shared multi-tenant pool. Questions pass through layered prompt-injection defenses before they reach your data, ingested content is fenced and scanned, secrets are encrypted at rest, and API keys are hashed and scoped. Deploy agent memory infrastructure with confidence.
- Isolated instance per customer — your own knowledge graph and store, not a shared tenant pool
- Layered prompt-injection defense: pattern detection, a dedicated classifier model, and streaming output filtering
- Every question screened by a prompt-injection classifier before it reaches retrieval
- Untrusted document, web, and tool content is fenced so planted instructions can't hijack the agent
- Ingested documents scanned and flagged for hidden injection attempts
- API keys hashed and scoped to collections (read or manage); connector & skill secrets encrypted at rest
- Per-key rate limiting, usage analytics, and key expiry / rotation
Key Metrics
Ask AI
Natural language Q&A for humans and agents alike
Ask questions about your knowledge base and get accurate, cited answers. Streaming responses, multi-step deep research, and conversation memory that persists across sessions — ready for both human users and agent pipelines.
- Natural language question answering
- Source citations with every answer
- Streaming responses for real-time UX
- Deep Research mode for multi-step reasoning — toggle on at any time
- Conversation memory across sessions (context curator)
Key Metrics
Frequently asked questions
Technical details about Cortex capabilities.
GraphRAG combines knowledge graph traversal with retrieval-augmented generation. Cortex automatically extracts entities and relationships from your documents, builds a traversable graph, and uses graph structure — communities, clusters, and multi-hop paths — alongside vector search to retrieve richer, more accurate context for LLM responses.
Give your agents real memory
See Cortex in action with our live demo, or register now for early access to the private beta.