Cognee — StartOS Package

Run Cognee — the open-source AI long-term memory platform — as a fully managed service on StartOS.

Cognee gives AI agents persistent memory across sessions. Ingest documents, emails, voice recordings, and notes — it builds a self-hosted knowledge graph with vector search, graph reasoning, and ontology generation.

This package wraps Cognee as a StartOS 0.4.0+ service with embedded databases, OpenRouter-ready configuration, and automated backup.

Detailed Setup & Usage Instructions →

---

Quick Start

Install via sideload on StartOS 0.4.0+:

# Download the latest build
curl -LO https://github.com/tylerkstevens/cognee-startos/releases/latest/download/cognee_x86_64.s9pk

# Install
start-cli package install --sideload cognee_x86_64.s9pk

Then use the Configure LLM action in the StartOS UI to set your API key, then start the service.

---

What's Inside

Component	Details
Upstream	Cognee — Apache 2.0 licensed
Container	cognee/cognee:latest (API) + ghcr.io/tylerkstevens/cognee-frontend:latest (UI)
Ports	3000 — native Cognee web UI (Open UI); 8000 — REST API
Databases	LanceDB (vectors), Kuzu (graph), SQLite (metadata) — all embedded, zero setup
LLM	OpenAI-compatible endpoint — pre-configured for OpenRouter
SDK	@start9labs/start-sdk v1.5.3

---

Features

• Native web UI — click Open UI to use Cognee's own interface on port 3000
• REST API — full control via HTTP at port 8000
• Self-hosted knowledge graphs — documents, emails, recordings, websites all become queryable memory
• Vector + graph search — semantic similarity and relationship traversal
• Multi-format ingestion — PDFs, text, CSV, images (captioning), audio (transcription), web pages
• Two-tier backup — binary tar (full restore) + semantic text exports (re-ingest safety net)
• Auto-ingestion pipelines — designed to integrate with Hermes Agent cron workflows (email, Pocket, Obsidian)

See the instructions.md for full API reference, dataset management, and troubleshooting.

---

Building Locally

git clone https://github.com/tylerkstevens/cognee-startos
cd cognee-startos
npm ci
npm run build              # compiles TypeScript → JavaScript via ncc

# Install system deps (first time only)
sudo apt-get install squashfs-tools squashfs-tools-ng

# Pack the s9pk
start-cli s9pk pack -o cognee_x86_64.s9pk --arch x86_64 --no-assets

Requires Node.js 22 and start-cli (download: https://github.com/Start9Labs/start-os/releases/download/v0.4.0-beta.9/start-cli_x86_64-linux).

CI/CD

Every push to master / main triggers a GitHub Actions workflow that:

1. Installs dependencies and compiles
2. Packs the .s9pk artifact
3. Creates a GitHub Release with tag build-{run_number}

Signed releases are available on the Releases page.

---

Registry Status

• ✅ Self-hosted registry — published to https://192.168.0.4:50286 (Personal Registry)
• 🔜 Community Registry submission — polish and submit for global StartOS discoverability

---

Project Structure

cognee-startos/
├── .github/workflows/build.yml   # CI/CD pipeline
├── startos/
│   ├── init.ts                    # First-run setup + config
│   ├── main.ts                    # Service lifecycle
│   ├── backups.ts                 # Backup/restore hooks (tar-based)
│   ├── actions/                   # Configurable actions (Configure LLM)
│   ├── manifest/
│   │   ├── index.ts               # Package manifest
│   │   └── i18n.ts                # Descriptions (en_US)
│   └── versions/                  # Version migration graph
├── icon.svg                       # Service icon
├── instructions.md                # User-facing documentation
└── package.json

---

License

• Package code: Apache-2.0
• Upstream Cognee: Apache 2.0

---

Links

• Upstream Cognee — the AI memory platform this packages
• Cognee Documentation
• Cognee Website
• StartOS SDK
• StartOS Registry Docs