Kraken is a self-hosted AI assistant you run on your own hardware. One instance handles coding, research, scheduling, browsing, and automation across every platform you connect. It builds a knowledge graph from every conversation, auto-creates reusable skills, and gets smarter the longer you use it.
from kraken import KrakenClient
client = KrakenClient(api_url="http://localhost:8080", api_key="sk-kraken-...")
# The agent remembers across sessions
client.chat("My name is Alice and I'm building a Rust compiler", session_key="alice")
# ... hours later ...
r = client.chat("What am I working on?", session_key="alice")
print(r.content) # "You're building a Rust compiler."
# Query the knowledge graph directly
results = client.memory.query("What do you know about Alice's projects?")
for entity in results.entities:
print(f"{entity.type}: {entity.name}")One assistant with one memory. Not another wrapper around a chat API.
Every conversation from every channel feeds the same knowledge graph. Your assistant knows your projects, preferences, and goals across Discord, Telegram, CLI, and anything else you connect.
Memory SystemA SOUL.md personality file you control. An auto-maintained user model that deepens over time. Your assistant learns how you think and what you care about.
Complex workflows become reusable skills automatically. The more you use Kraken for real tasks, the better it gets at helping you with similar tasks in the future.
SkillsDrop-in /v1/chat/completions endpoint. Works with any OpenAI client library. Session routing and memory ride alongside the standard API.
Type-safe client for chat, memory, skills, sessions, and identity. Streaming and session routing out of the box.
Python SDKRuns entirely on your hardware. Docker-isolated code execution with networking and port forwarding. Browser automation with SSRF protection. Your data never leaves your infrastructure.
ArchitectureYour App (Python SDK / any HTTP client)
│
▼
┌─────────────────────────────────────┐
│ Kraken API (Hono) │
│ REST + WebSocket + streaming │
│ OpenAI-compatible /v1/chat/* │
└──────────────┬──────────────────────┘
│
┌──────────┼──────────────────────┐
│ │ │
▼ ▼ ▼
Worker PostgreSQL 17 Neo4j 5
(BullMQ) + pgvector (knowledge graph)
│ │ │
├── Memory extraction ├── Entities
├── Community detection ├── Relationships
├── User model updates └── Communities
├── Skill reflection
├── Dream cycle ┌─────────────┐
└── Schedule execution │ Redis 7 │
│ (queues) │
┌─────────────┐ └─────────────┘
│ Chromium │
│ (browser) │ ┌─────────────┐
└─────────────┘ │ Sandbox │
│ (Docker) │
└─────────────┘