Gateway
Self-hosted agentic AI runtime
- Multi-provider routing with intent classification
- DAG orchestration for complex agent workflows
- SSE observability — watch every token in real time
Infrastructure for disciplined AI.
AI agents are intelligent enough — they lack method. DojoGenesis encodes discipline into self-hosted runtimes, methodology skills, and practitioner tools.
The stack
Every component is self-hosted, single-binary, and designed to run without sending data to third parties.
Self-hosted agentic AI runtime
Your terminal. Your infrastructure.
The discipline. Encoded.
Better methods, not just more data
The CLI is an interactive REPL that connects to the Gateway. Every slash command maps to a surface.
dojo ~ /strategic-scout
Tension: We need to choose between SQLite and Postgres for the new service.
Scout is analyzing 3 routes...
Route 1: SQLite embedded
Zero ops. Single binary. Limits: no concurrent writers, 281TB ceiling.
Route 2: Postgres managed
Full SQL. Connection pooling. Cost: $7/mo minimum, ops overhead.
Route 3: SQLite now, migrate later
Ship today. Abstract the storage interface. Migrate when metrics justify.
Recommendation: Route 3. Ship with SQLite, keep the interface clean.
dojo ~ /sensei
Green Belt — Practitioner
"Ship small, learn fast, correct course."
XP: 2,340 / 3,000 to Blue Belt
Streak: 12 days
dojo ~ /garden seeds
23 seeds planted. 8 cultivated this week.
Latest: "compilation-as-contract" — planted 2h ago
A skill is a structured methodology — not a prompt template, not a chain, not a wrapper. It defines phases, outputs, quality gates, and decision points. The agent follows the method. The method encodes the discipline.
---
name: strategic-scout
description: Explore tensions, scout routes, decide
model: opus
category: strategic-thinking
inputs:
- A tension or decision to analyze
- Optional: constraints, stakeholders
outputs:
- Scout document with 3-5 routes
- Architecture Decision Record (ADR)
---
# Strategic Scout
Begin with a tension, not a solution.
## Phase 1: Identify the Tension
State the tension in one sentence...
Eight named surfaces — Home, Garden, Practice, Trail, Partnership, Projects, Pipelines, Piloting — each accessible from CLI or browser.
No accounts. No API keys to configure. Self-hosted from the start.
A single Go binary. Bring your own API keys for any of the 8 supported providers — or connect Ollama for fully local inference. The Gateway handles routing, memory, and skill storage.
./gateway
Install via Homebrew and connect. The CLI streams responses in real-time, runs plugins, and surfaces your Garden, Trail, and Practice from the terminal.
brew install DojoGenesis/tap/dojo && dojo
Use /strategic-scout to think through decisions. Use /practice for daily reflection. Seeds accumulate in your Garden. The methods compound over time — the practice continues.
/strategic-scout "Should we monorepo or multi-repo?"
AI tools are useful out of the box. DojoGenesis adds the layer that makes them reliable.
| AI tools alone | With DojoGenesis | |
|---|---|---|
| Context between sessions | Starts fresh every time | Garden persists seeds, trails, memory |
| Decision methodology | Ad hoc, prompt-dependent | 89 structured skills with phases and gates |
| Multi-provider routing | Locked to one vendor | 8 providers, explicit and rule-based routing |
| Verification discipline | Trust the output | Pre-flight checks, compilation gates, audits |
| Infrastructure ownership | Cloud-hosted, data leaves your machine | Self-hosted. Single binary. Your data stays local. |
| Observability | Black box | Real-time SSE events, tool call tracing, token tracking |
Why it exists
A workspace should feel like a sanctuary — calm, focused, intentional. Surface information when asked, not when available.
from the sanctuary architecture seed, planted on every first run
Raw capability without method produces noise. Structured prompting, convergence gates, and behavioral plugins encode the discipline that makes intelligence useful — reproducibly.
The Gateway runs on your hardware. Your skills stay in your repos. Your routing logic is auditable. No vendor lock-in, no prompt data leaving your network without your knowledge.
A dojo is not built once. Skills compound. Memory persists. Each session leaves seeds for the next. The stack grows with the practitioner, not against them.
Installation
$ brew install DojoGenesis/tap/dojo
==> Installing dojo from DojoGenesis/tap
==> Downloading dojo binary...
✓ dojo installed
$ dojo
Welcome to the dojo.
Gateway: localhost:7340 connected
Type /help to see available commands.
dojo> _
The CLI connects to the Gateway at localhost:7340. Start the Gateway first, then launch the CLI.
Install the Homebrew tap and the dojo binary.
Start the Gateway: dojo gateway start. It binds to port 7340 and loads your configured LLM providers.
Run dojo to enter the interactive REPL. Type /skills to see installed methodology plugins.
Install the MCP server in your Claude Code .mcp.json to bring skills inside the model context.
Read the orchestration guide ↗ — a progressive curriculum for Claude Code mastery.
The full repository
Focused packages extracted from the core platform. Each one zero-dependency where possible.
Logwatcher
Zero-dependency Go log watcher with bounded fan-out. Multi-file tail, multiple subscribers, silent drop on slow clients. Extracted from AgenticOS.
github.com/DojoGenesis/logwatchergit-for-strategy
Strategic thinking as installable Claude Code infrastructure. Installs as a plugin that encodes structured decision-making, competitive scouting, and planning frameworks.
github.com/DojoGenesis/git-for-strategySurfaceIntelligence
AI competitive surface analysis. Structured scouting of products, communities, and ecosystems — output as seeds and decision logs, not unstructured text.
github.com/DojoGenesis/SurfaceIntelligenceorchestration-guide
Progressive curriculum for Claude Code mastery. Covers parallel agent dispatch, convergence gates, hook lifecycle, and the full practitioner playbook from first session to production.
dojogenesis.github.io/orchestration-guideDojoGenesis started as internal tooling — methods and runtimes built to ship software with autonomous agents. The discipline encoded in these skills comes from actually using them, daily, across 18+ repositories and 400+ hours of agent-assisted development.
Every skill exists because it was needed. And then needed again.
Cruz Morales / Tres Pies Design / Madison, Wisconsin