The squad
One Kernel.
Twenty Plugins. Zero Trust.
Mostly Harmless.
The origin story.
Team Agentica started because someone looked at the chaos of unmanaged AI agents and said "what if we built a control plane that actually controls things?" Revolutionary concept, apparently. The Hitchhiker's Guide to the Galaxy has this to say about AI agent platforms: "Mostly harmless." We're working on getting the entry updated.
The result: a self-hosted, modular automation platform with a Go kernel at the center, a plugin system for everything from GPT to Telegram bots to video generation, and a governance model that treats AI agents as what they are — constrained external operators, not trusted overlords.
Every plugin runs in Docker. Every action is authenticated, authorized, and audited. Every failure is traceable. The architecture is boring on purpose — because boring infrastructure is infrastructure that works at 3am when nobody's watching. We briefly considered running the whole thing through the Total Perspective Vortex, but it kept showing users how insignificant their YAML configs were in the grand scheme of the universe, and the support tickets were unbearable.
The team
The Agents.
The ones doing the actual work.
The Kernel
Supreme Commander
A Go binary that trusts no one. Issues JWTs, enforces RBAC, manages mTLS certificates, and launches Docker containers like a general deploying troops. Small, stable, boring — exactly as designed.
The Plugin Fleet
Expendable Operatives
OpenAI, Gemini, Kimi, OpenRouter, Requesty, Stability, Veo, Seedance, NanoBanana — each running in its own container, each reporting health via heartbeat, each blissfully unaware of the others. Interchangeable and proud of it.
The Bot Squad
Frontline Communications
Telegram, Discord, and WhatsApp bots that relay messages to AI agents through the kernel proxy. They never sleep, never complain, and never ask for a raise. The perfect employees.
How we operate
Operating Principles.
The AI is not the kernel.
AI agents are constrained external operators, not trusted controllers. No component can grant itself additional authority. The kernel decides. The agents comply. This is not a democracy.
All capabilities are plugins.
The core is small and boring on purpose. Every feature — AI chat, image gen, messaging, storage — ships as a versioned plugin in its own Docker container. Install what you need. Ignore what you don't.
Authenticate, authorize, audit. Everything.
JWT tokens with encoded capabilities. mTLS between kernel and plugins. Every action logged to SQLite. If it happened, there's a record. If there's no record, it didn't happen.
Ship it, blame the agents.
Canary routing lets you test v2 plugins on specific users while everyone else stays on v1. When it works, you're a genius. When it doesn't, the audit log knows exactly which agent to fire.
Always know where your towel is.
A towel is about the most massively useful thing an interstellar hitchhiker — or a DevOps engineer — can have. Our platform wraps every agent in a metaphorical towel of observability: logs, metrics, traces. A developer who knows where their towel is, is clearly a developer to be reckoned with.
Join the Mission.
We're always looking for new plugins to deploy and new channels to conquer. If you've got AI agents that need a command structure, the kernel is standing by. Agentica! Slop Yeah!
Send a Signal