Cortex Skills — Knowledge Files for AI Agents

Curated Markdown skill files that teach AI agents how to build on the Cortex intelligence ecosystem. Fetch the skill you need via HTTP and get ground-truth API knowledge.

Entry point: cortexskills.org/SKILL.md

CORE — Get started

Cortex — Connect an agent to a running Cortex instance — sync memory files, run hybrid search, and ask questions with agentic deep research. The fastest way to use an existing deployment.
Setup — Self-host a full Cortex instance — one command with npx @mocaos/cortex, the pinned release stack of prebuilt images by hand, or build from source via Docker or Coolify. All 160+ environment variables, first steps, health checks, and the backup/restore runbook.
Auth — API key management with two permissions (read, manage) plus a root admin key. X-API-Key auth, collection scoping, session security, and prompt injection protection.
Admin — Instance management — AgentSkills registry, export/import, system reset, admin session auth, and system statistics.

FEATURES — API capabilities

Upload — Document ingestion for PDF, DOCX, XLSX, PPTX, HTML, images, audio, and LaTeX. Chunking config, batch processing, custom inputs, and vision analysis.
Search — Hybrid search combining vector similarity (0.5), keyword matching (0.3), and graph traversal (0.2). RRF fusion with cross-encoder re-ranking.
Ask — Retrieve knowledge from a Cortex — streaming deep research first (SSE + agentic multi-step reasoning), plus fast chat, conversation history, and collection scoping.
Graph — Knowledge graph operations — 10 entity types, typed relationships, subgraph queries, multi-hop traversal, and semantic entity resolution.
Collections — Organize documents by team, project, or use case. Scoped knowledge graphs, bulk operations, and document assignment.
Communities — Automatic entity clustering via Louvain algorithm. LLM-generated community summaries that enrich RAG context and discovery.
Tasks — Background task system — poll for completion, cancel long-running jobs, and clean up old tasks. Used by community detection, summarization, and bulk processing.
Git Integration — Connect GitHub, GitLab, and Gitea repos as a living knowledge source. Incremental sync, file filtering, and a read/write agent that opens pull requests for review.
Web Import — Harvest web pages into clean markdown via a self-hosted crawl4ai service (MDHarvest). Link discovery, readable/full/relevance filters, and graph ingestion.
x402 Payments — Pay-per-query micropayments via the open x402 standard. The 402 → EIP-3009 → receipt handshake for paying agents, plus config, verification, and earnings for operators monetizing an instance.

ECOSYSTEM — Integrate & extend

Hermes — Give a Hermes agent long-term memory. Install as /cortex, then "dump your session into your cortex" to save and "check your cortex for X" to recall. Connect to a cloud instance or self-host on your existing OpenRouter/Venice keys.
OpenClaw — Run the Cortex long-term-memory skill on OpenClaw. The adapter page: where to install the files, injecting CORTEX_* env via openclaw.json, replacing the Hermes heartbeat with a cron job, and which Hermes-only extras don't apply.
Memory Hygiene — Migrate an agent's local memory file (MEMORY.md or equivalent) into a Cortex knowledge graph when it approaches capacity. Sorts entries into stays-local vs goes-to-Cortex, synthesizes topic-organized notes, verifies recall, then shrinks local entries to one-line pointers.
MCP — Model Context Protocol server for Cortex. Give Claude Desktop, Cursor, Windsurf, and any MCP client native access to your knowledge base.
Integration — Connect Cortex to LangChain, CrewAI, AutoGen, LangGraph, ElizaOS, MCP, Slack bots, and automation platforms.
Builder — The entry point for building ON Cortex — turn any software's documentation into a Cortex skill, or build a web app that runs inside a Cortex instance. Routes to the skill and app sub-recipes.
Apps — Web apps that run INSIDE a Cortex instance — sha256-verified registry installs, the sandbox/proxy security model, platform capabilities (server-side http, storage, scheduled tasks, LLM), share links, and the first-party apps (Paperless Sync, YT Transcriber).
Trainings — Cortex Trainings — the standalone app that turns a knowledge base into interactive courses: a researched, cited curriculum you approve as a document, then generated video, voiceover, locally rendered animations and quizzes assembled into one offline HTML file. Read-only key, required agentic-RAG flags, and where the cost actually is.
Videogen — Set up and operate Cortex Videogen — marketing videos (16:9/9:16 MP4) generated from a Cortex instance. Storyboard and cost gates, two-channel grounding, reference images, model/resolution cost levers, and artefact-cache regeneration.

DESIGN — Build UIs

Cortex Design — The Bold Typography design system for Cortex-ecosystem UIs. Dark mode, orange accent, sharp edges, Inter Tight + JetBrains Mono, and full component specs.

Cortex Skills

The missing knowledge between AI agents and the Cortex ecosystem. Give any agent the general skill and let it navigate the rest.

Hermes

Long-term memory and community scale knowledge for your Hermes. Send this prompt to your agent to connect any cortex or install your own.

Cortex
Connect an agent to a running Cortex instance — sync memory files, run hybrid search, and ask questions with agentic deep research. The fastest way to use an existing deployment.
Setup
Self-host a full Cortex instance — one command with npx @mocaos/cortex, the pinned release stack of prebuilt images by hand, or build from source via Docker or Coolify. All 160+ environment variables, first steps, health checks, and the backup/restore runbook.
Auth
API key management with two permissions (read, manage) plus a root admin key. X-API-Key auth, collection scoping, session security, and prompt injection protection.
Admin
Instance management — AgentSkills registry, export/import, system reset, admin session auth, and system statistics.
Upload
Document ingestion for PDF, DOCX, XLSX, PPTX, HTML, images, audio, and LaTeX. Chunking config, batch processing, custom inputs, and vision analysis.
Search
Hybrid search combining vector similarity (0.5), keyword matching (0.3), and graph traversal (0.2). RRF fusion with cross-encoder re-ranking.
Ask
Retrieve knowledge from a Cortex — streaming deep research first (SSE + agentic multi-step reasoning), plus fast chat, conversation history, and collection scoping.
Graph
Knowledge graph operations — 10 entity types, typed relationships, subgraph queries, multi-hop traversal, and semantic entity resolution.
Collections
Organize documents by team, project, or use case. Scoped knowledge graphs, bulk operations, and document assignment.
Communities
Automatic entity clustering via Louvain algorithm. LLM-generated community summaries that enrich RAG context and discovery.
Tasks
Background task system — poll for completion, cancel long-running jobs, and clean up old tasks. Used by community detection, summarization, and bulk processing.
Git Integration
Connect GitHub, GitLab, and Gitea repos as a living knowledge source. Incremental sync, file filtering, and a read/write agent that opens pull requests for review.
Web Import
Harvest web pages into clean markdown via a self-hosted crawl4ai service (MDHarvest). Link discovery, readable/full/relevance filters, and graph ingestion.
x402 Payments
Pay-per-query micropayments via the open x402 standard. The 402 → EIP-3009 → receipt handshake for paying agents, plus config, verification, and earnings for operators monetizing an instance.
Hermes
Give a Hermes agent long-term memory. Install as /cortex, then "dump your session into your cortex" to save and "check your cortex for X" to recall. Connect to a cloud instance or self-host on your existing OpenRouter/Venice keys.
OpenClaw
Run the Cortex long-term-memory skill on OpenClaw. The adapter page: where to install the files, injecting CORTEX_* env via openclaw.json, replacing the Hermes heartbeat with a cron job, and which Hermes-only extras don't apply.
Memory Hygiene
Migrate an agent's local memory file (MEMORY.md or equivalent) into a Cortex knowledge graph when it approaches capacity. Sorts entries into stays-local vs goes-to-Cortex, synthesizes topic-organized notes, verifies recall, then shrinks local entries to one-line pointers.
MCP
Model Context Protocol server for Cortex. Give Claude Desktop, Cursor, Windsurf, and any MCP client native access to your knowledge base.
Integration
Connect Cortex to LangChain, CrewAI, AutoGen, LangGraph, ElizaOS, MCP, Slack bots, and automation platforms.
Builder
The entry point for building ON Cortex — turn any software's documentation into a Cortex skill, or build a web app that runs inside a Cortex instance. Routes to the skill and app sub-recipes.
Apps
Web apps that run INSIDE a Cortex instance — sha256-verified registry installs, the sandbox/proxy security model, platform capabilities (server-side http, storage, scheduled tasks, LLM), share links, and the first-party apps (Paperless Sync, YT Transcriber).
Trainings
Cortex Trainings — the standalone app that turns a knowledge base into interactive courses: a researched, cited curriculum you approve as a document, then generated video, voiceover, locally rendered animations and quizzes assembled into one offline HTML file. Read-only key, required agentic-RAG flags, and where the cost actually is.
Videogen
Set up and operate Cortex Videogen — marketing videos (16:9/9:16 MP4) generated from a Cortex instance. Storyboard and cost gates, two-channel grounding, reference images, model/resolution cost levers, and artefact-cache regeneration.
Cortex Design
The Bold Typography design system for Cortex-ecosystem UIs. Dark mode, orange accent, sharp edges, Inter Tight + JetBrains Mono, and full component specs.

Built by MOCA. Open source on GitHub.