Generate One platform

Shell 59.4%
Python 39.7%
Dockerfile 0.9%

Find a file

g1admin 94f4ccc8d3 Some checks failed CI / smoke-test (push) Failing after 35s Details Commit Summary / summarize (push) Failing after 31s Details Secret Scan / scan (push) Failing after 36s Details docs: premium README overhaul ✨		2026-03-22 06:12:07 +00:00
.forgejo/workflows	COD-1: Automated weekly version drift detection workflow (#92 )	2026-03-17 06:46:27 +00:00
.gitea/issue_template	chore: add infrastructure change issue template	2026-03-04 13:47:40 +00:00
docs	docs: Lab/Run v10 — bookend protocol (Decision f3b36a8d)	2026-02-28 04:36:18 +00:00
infrastructure	COD-1: Automated weekly version drift detection workflow (#92 )	2026-03-17 06:46:27 +00:00
src/openproject-mcp	fix(openproject-mcp): shorten 2 tool names exceeding 64-char MCP limit	2026-02-24 20:49:07 +00:00
test	cleanup: remove test file from 12.2.3	2026-02-17 23:50:09 +00:00
.gitignore	Add .gitignore	2026-02-25 15:39:16 +00:00
.gitleaks.toml	Add gitleaks config for secret scanning CI	2026-02-25 20:38:37 +00:00
CLAUDE.md	docs: add Rules 64 (CD/CC execution split) and 65 (stale container labels)	2026-03-21 00:55:42 +00:00
LICENSE	Add MIT LICENSE	2026-02-25 15:39:36 +00:00
PROJECT-STATE.md	chore: freeze PROJECT-STATE.md — migrated to Directus (Dispatch 52fb3a10)	2026-03-04 15:44:09 +00:00
README.md	docs: premium README overhaul ✨	2026-03-22 06:12:07 +00:00

README.md

Generate One 🛸

AI infrastructure that answers to you. Self-hosted, sovereign, and GPU-accelerated on a single RTX 5090 server.

✨ Overview

Generate One is a full-stack AI platform running on a single dedicated server (RTX 5090 GPU, 32 GB VRAM). It provides LLM routing across 8 model tiers, graph-based persistent memory, vector search with 4-stream Reciprocal Rank Fusion, and 195+ tools exposed via the Model Context Protocol (MCP). All services are orchestrated through Coolify and routed via Traefik with Let's Encrypt TLS.

This is the monorepo root — it holds the infrastructure manifest, version tracking, CI/CD workflows, and cross-stack documentation. Individual service stacks live in their own repos.

🏗️ Architecture

┌──────────────────────────────────────────────────────────┐
│                        Clients                           │
│              Claude Code · Chat UI                       │
└──────────────────────┬───────────────────────────────────┘
                       │
          ┌────────────▼─────────────┐
          │   Traefik + Let's Encrypt │
          │   (reverse proxy / TLS)   │
          └────┬──────────┬──────────┘
               │          │          │
   ┌───────────▼──┐  ┌────▼─────┐  ┌▼────────────┐
   │   g1-api     │  │  g1-gpt  │  │   g1-mcp    │
   │   Fusio      │  │LibreChat │  │  MetaMCP    │
   │  REST API    │  │ Chat UI  │  │ 195+ tools  │
   └──────────────┘  └──────────┘  └─────────────┘
                       │
   ┌───────────────────▼──────────────────────────────┐
   │                   g1-llm                         │
   │  LiteLLM (8 tiers) · vLLM (GPU) · Qdrant        │
   └──────────────────────┬───────────────────────────┘
                          │
   ┌───────────────────────▼──────────────────────────┐
   │                   g1-brain                       │
   │  Neo4j · Graphiti · knowledge-mcp                │
   │  FalkorDB · reasoning-tools                      │
   └──────────────────────────────────────────────────┘

📦 Stacks

Stack	Domain	Purpose
g1-core	—	Shared PostgreSQL 16 + Valkey 9.0.1
g1-llm	llm.generate.one	LiteLLM proxy, 3× vLLM (GPU), Qdrant
g1-brain	memory.generate.one	Neo4j, Graphiti, knowledge-mcp (4-stream RRF)
g1-mcp	mcp.generate.one	MetaMCP gateway, 9 namespaces, ~195 tools
g1-frontend	app.generate.one	Next.js apps: web, app, admin
g1-agent-backend	agent.generate.one	PydanticAI streaming agent API
g1-web	—	SearXNG metasearch, Crawl4AI scraping
g1-gpt	gpt.generate.one	LibreChat v0.8.2, MongoDB
g1-state	cms.generate.one	Directus CMS, Infisical secrets
g1-auth	auth.generate.one	Authentik SSO (12 OIDC providers)
g1-observe	observe.generate.one	Langfuse v3 LLM observability
g1-automation	n8n.generate.one	n8n workflow automation
g1-git	git.generate.one	Forgejo v14, Actions runner

⚡ MCP Namespaces

Namespace	Tools	Backend
g1-brain	43	graphiti-mcp, knowledge-mcp, reasoning-tools, logic-lm
g1-code	112	gitea-mcp, section-mcp
g1-web	~15	searxng-mcp, crawl4ai
g1-math	~15	sympy-mcp
g1-time	5	time-mcp
g1-presenter	~5	presenter-mcp
g1-project	111	worklenz-mcp

🔧 Model Tiers

Tier	Primary Model	Provider
`g1-llm`	K2-instruct	Groq → OpenRouter
`g1-llm-turbo`	GPT-OSS-120B	Groq → Cerebras → OpenRouter
`g1-llm-mini`	Qwen3-235B	Cerebras → OpenRouter
`g1-llm-code`	GLM-4.7	Cerebras → OpenRouter
`g1-llm-pro`	K2.5	OpenRouter
`g1-llm-micro`	Qwen3-8B-FP8	Local vLLM → Cerebras
`g1-vlm`	Qwen3-VL-30B	OpenRouter

🖥️ Infrastructure

Component	Value
Server	au2404 / 161.184.162.156
GPU	RTX 5090, 32 GB VRAM
GPU Budget	90% VRAM (embedding 30% + reranker 15% + small model 45%)
Orchestration	Coolify (cloud-hosted)
Reverse Proxy	Traefik v3 + Let's Encrypt
Monitoring	Gatus (uptime) · Beszel (resources) · Langfuse (LLM quality)

🔖 Version Manifest

Service versions are tracked in infrastructure/versions.yaml and validated weekly by the version-check CI workflow. Key pinned versions:

Service	Version
PostgreSQL	16
Valkey	9.0.1
LiteLLM	1.81.12
Qdrant	v1.17.0
vLLM	v0.16.0
Neo4j	5.26-community
MetaMCP	2.4.22
Forgejo	14.0.2
Langfuse	v3.158.0
Authentik	2026.2.0

📚 Documentation

Project documentation lives in Directus CMS — the canonical source of truth for skills, dispatches, decisions, and reports. Each stack repo contains its own compose files and README.

🛡️ Part of Generate One

Generate One — AI infrastructure that answers to you.

Self-hosted, sovereign AI platform. generate.one

Licensed under AGPL-3.0.

README.md Unescape Escape