ODIN
A personal AI command center. Six specialised agents, one orchestrator, and no team required.
→ Follow the build in public on LinkedInSolo founders don’t have a team. They have a task list that never ends and a single brain trying to cover every role at once: engineer, product manager, analyst, marketer, researcher. Context switches constantly. Things fall through.
ODIN is my answer to that. Not a chatbot. Not a workflow tool. A command center: a persistent, memory-equipped orchestration layer that routes work to the right specialist agent, tracks what is in flight, and surfaces the result when it is done. I built it for myself, in public, while building Clarai.
// 01 · THE PROBLEM
A solo founder is a broken team.
When you do everything alone, you don’t just lose time. You lose context. You start a marketing task in the headspace of an engineer. You make a product decision with zero user data. You build things nobody asked for, because nobody was around to ask the uncomfortable questions.
I didn’t need an assistant that could answer questions. I needed a team that could hold their own context, challenge my assumptions, and do the work without me holding their hand.
// 02 · THE SYSTEM
Six agents. One orchestrator. No redundancy.
Every agent in ODIN has a defined specialty and a mandate to push back.
ODIN
Orchestrator
The orchestrator and external brain. He carries context across sessions, connects dots, routes incoming work to the right agent, and synthesises results. Not a chatbot. The All-Father.
LUCA
Senior developer
A senior developer. He knows the codebase, has opinions about architecture, and tells me when something is going to hurt in three months. He creates real GitHub repositories, writes production code, and opens pull requests.
MAYA
Product owner
The product owner. Her job is to protect me from myself. Before anything gets built, Maya asks why: who needs this, what is the smallest version that proves the point, and what are we assuming that we haven't validated.
NOVA
Internal analyst
The internal analyst. She works with data I already have: outreach conversion rates, runway calculations, agent usage patterns. She doesn't present five options. She gives one recommendation with reasoning.
FREYA
Marketing & sales
The marketing and sales coach. She designs outreach campaigns, writes LinkedIn DMs in my voice, prepares me for objections, and closes the loop when NOVA analyses the results.
MIMIR
External researcher
The external researcher. She sweeps the market for signals: competitors, pricing, technology trends, what is being built and what has failed. Her output becomes NOVA's input.
// 03 · HOW IT WORKS
Route. Dispatch. Track. Surface.
A task comes in. ODIN decides who owns it. The agent works asynchronously, in the background while I continue, and ODIN surfaces the result when it is ready. If another agent should pick up from there, ODIN offers the handoff. It never auto-chains without asking.
Every agent has a voice. ElevenLabs handles text-to-speech, and wake-word detection routes spoken requests to the right agent without touching a keyboard.
// 04 · WHAT I BUILT
A Next.js PWA with a 3D neural UI and real tool use.
Next.js 14 App Router, TypeScript and Tailwind on the front, the Anthropic API underneath (Claude Haiku for routing, Sonnet for the heavy work), ElevenLabs for voice, GitHub Octokit for LUCA’s code operations, and the Web Speech API for wake-word detection.
The interface is a 3D rotating neural network rendered in WebGL. Neurons fire, thought-pulses travel the synapses, and voice input drives the animation in real time via Web Audio RMS. Every agent has a colour, and the network lights up in that colour when the agent is active.
Under the hood: file-based memory per project, a cost tracker that enforces a daily budget per model, prompt caching on static system prefixes, and a confirmation gate before LUCA pushes anything to a remote repository. Agents are configuration-driven, not hardcoded. New agents load from .claude/agents/ without a restart.
// 05 · WHAT THIS IS, REALLY
The team structure is the product.
The interesting part of ODIN isn’t the code. It is the agent design: the decision to give every specialist a mandate to challenge, a scope that doesn’t bleed into the next agent’s lane, and a clear answer to “what does this agent say no to?”
Most AI tooling gives you one model and a text box. ODIN gives you a team with defined roles, institutional memory, and the ability to say “that’s not my department.” I built it in the open while using it to build Clarai, with every agent system prompt and architectural decision documented publicly.
// TECHNICAL DETAILS
The stack.
Building in public · 2026