📌 This repository is a snapshot, not the living tree
The application code here was synced from the development tree on 8 May 2026; only
documentation and CLI packaging changed after that, up to 25 May 2026. It is a
faithful copy of what ZenAI was on that date — readable, citable and runnable — but it
is not the current state of development, and it does not receive fixes.
The CI pipeline is set to manual dispatch for the same reason: it tests files that the
development tree has since removed, so an automatic run reported a failure that nobody
here could act on. That says nothing about the snapshot itself.
What is current: the memory library ZenBrain
is actively developed and released to npm, and the research record behind both is kept
up to date at zensation.ai/en/publikationen.
The terminal agent in cli/ was published as @zensation/cli.
It is retired: every version is deprecated on npm (1 October 2026), because it reads
API_URL and VITE_API_KEY from the environment as undocumented fallbacks. Maintained
memory tooling: @zensation/mcp.
Questions about the state of ZenAI: [email protected]
What is ZenAI?
ZenAI is a self-hosted AI platform with neuroscience-inspired memory. Unlike ChatGPT or Claude, ZenAI remembers your conversations, learns from your preferences, and improves over time — using the same mechanisms your brain uses.
Key Features
- 60 AI Tools across 14 categories (memory, web search, code execution, GitHub, maps, email, documents, and more)
- 7-Layer Memory (Working, Short-Term, Episodic, Semantic, Procedural, Core, Cross-Context) powered by ZenBrain
- FSRS Spaced Repetition — your AI never forgets what matters
- 4 Context Isolation — separate memory for personal, work, learning, and creative
- Multi-Agent System — researcher, writer, reviewer, coder agents with debate protocol
- RAG Pipeline — HyDE + Cross-Encoder reranking + GraphRAG hybrid retrieval
- Real-Time Voice — WebSocket STT/TTS pipeline with turn-taking
- Extended Thinking — visible AI reasoning with budget management
- Sleep-Time Compute — background memory consolidation and contradiction detection
- MCP Ecosystem — connect external tools via Model Context Protocol
Architecture
Frontend: React + TypeScript (Vite)
Backend: Express.js + TypeScript
AI: Claude API (primary), Mistral (fallback), Ollama (local)
Database: PostgreSQL + pgvector (4 isolated schemas)
Memory: ZenBrain 7-layer architecture
Cache: Redis (optional)
Quick Start
Prerequisites
- Node.js 20 (this snapshot declares
"node": "20.x")
- PostgreSQL 15+ with pgvector extension
- Anthropic API key
Setup
# Clone
git clone https://github.com/zensation-ai/zenai.git
cd zenai
# Backend
cd backend
cp .env.example .env
# Edit .env with your DATABASE_URL and ANTHROPIC_API_KEY
npm install
npm run build
npm run dev
# Frontend (new terminal)
cd frontend
npm install
npm run dev
Open http://localhost:5173 in your browser.
Docker Compose
Two compose files ship with this snapshot.
# Self-hosting: Postgres (pgvector) + Redis + backend on :3000
cp .env.docker .env
docker compose up -d
# Infrastructure only — Postgres + Redis, run backend and frontend locally
docker compose -f docker-compose.dev.yml up -d
The frontend is not part of the compose stack; run it as shown under
Setup.
API
ZenAI exposes 250+ API endpoints. Key areas:
| Area | Endpoints | Description |
|---|
| Chat | /api/chat/* | Streaming chat with tool use, vision, artifacts |
| Memory | /api/:context/memory/* | FSRS review, procedures, hybrid search |
| Ideas | /api/:context/ideas/* | CRUD with semantic search, topics, drafts |
| Tasks | /api/:context/tasks/* | Kanban, Gantt, dependencies, projects |
| Email | /api/:context/emails/* | Send/receive with AI categorization |
| Agents | /api/agents/* | Multi-agent execution, templates, streaming |
| Knowledge Graph | /api/:context/graphrag/* | Entity extraction, hybrid retrieval |
| Voice | /api/:context/voice/* | Real-time STT/TTS pipeline |
Full API documentation: /api-docs (Swagger UI when running)
Testing
# Backend (~11,500+ tests)
cd backend && npm test
# Frontend (~1,400+ tests)
cd frontend && npx vitest run
# CLI (108 tests)
cd cli && npm test
Memory System
ZenAI's memory is powered by ZenBrain — the neuroscience-inspired memory system:
| Layer | Inspired By | Function |
|---|
| Working Memory | Baddeley (1974) | Active task focus (7±2 items) |
| Short-Term | Session context | Current conversation |
| Episodic | Tulving (1972) | Concrete experiences |
| Semantic | FSRS + Ebbinghaus | Facts with spaced repetition |
| Procedural | Cognitive psychology | "How to do X" skills |
| Core | Letta/MemGPT | Pinned, user-editable facts |
| Cross-Context | ZenBrain architecture | Shared knowledge across domains |
Paper & Citation
The memory architecture underneath ZenAI is documented and peer-reviewable:
@article{bering2026zenbrain,
title = {ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture for Autonomous AI Systems},
author = {Bering, Alexander},
year = {2026},
eprint = {2604.23878},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2604.23878}
}
To cite this snapshot itself rather than the architecture, see CITATION.cff.
Contributing
See CONTRIBUTING.md for guidelines.
License
Apache 2.0
