Pantheon - AI Orchestration Platform
Completed: Present
An agentic AI orchestration platform, built to run my own fractional CTO delivery work at scale — and being open-sourced piece by piece as each component proves itself.
The Problem
Delivering fractional CTO work across multiple concurrent engagements means constant context-switching between clients, stacks, and priorities. Off-the-shelf AI coding tools handle single-session tasks well, but there was no toolable way to route work across models/providers by health and availability, broker credentials safely across engagements, or run multi-step build→test→review→ship pipelines without a human babysitting every step.
The Solution
Architected "Pantheon" — a modular multi-agent orchestration system made of independently-usable, named components: health-aware LLM/lane routing (Heimdall), a harness-side credential broker that keeps secrets out of LLM context (Portunus), an ideation-and-decision surface for the human-in-the-loop review cycle (Consus), swarm orchestration tooling (Dostal Swarm), and the orchestration layer tying it together (Dostal Orchestration). Each component works standalone or plugs into the full platform, and is being open-sourced independently as it matures rather than held back waiting for the whole system to be "done."
The Impact
First public release: a multi-agent workflow orchestrator for Claude Code (plan, build, test, and review software with coordinated AI teams, cross-model execution) — the same class of tooling that ran this site's own engineering work. Additional Pantheon components (health-aware routing, credential brokering, review dashboards) are rolling out publicly starting August 2026. Built and used daily on real delivery work, not a demo.