Corellia
Corellia is a vendor-neutral control plane for AI agents — a single platform to spawn, deploy, govern, and manage agents at scale across any model, provider, or agent framework.
Core trick
A harness interface spec lets any agent framework plug into one Go control plane that enforces tool-scope and model policy per agent.
Description
Corellia is a vendor-neutral control plane for AI agents — a single platform to spawn, deploy, govern, and manage agents at scale across any model, provider, or agent framework.
The problem:
As organisations begin deploying AI agents across their workforce, they hit a fragmentation wall fast. Deployment, tool permissions, observability, and orchestration are each solved by different point tools. None talk to each other, and most lock you into a specific stack.
The ones that do always result in model and/or model vendor lock-in (e.g. AWS, Azure etc.)
What Corellia does:
Admins define the guardrails — which tools an agent can access, what model it runs, where it deploys — and Corellia enforces them. Think AWS IAM, but for agents.
The v1 flow:
Pick a harness from the catalog, give your agent a name and model, equip it with tools and scopes, configure its deployment, and launch — one Fly.io machine per agent, secrets-isolated, auto stop/start, pinned to an immutable Docker image digest so governance is auditable and reproducible.
Under the hood:
Go + Connect-go backend, Next.js 15 frontend, Proto IDL as the single FE↔BE contract, sqlc for typed SQL, Supabase auth with offline JWKS validation. The harness interface contract — how Corellia talks to any agent framework — is designed like the LSP: define the spec once, every new framework becomes a one-time integration.
v1 ships with:
- Hermes Agent (Nous Research) as the first harness
- Tool-scope governance (URL / command / path allowlists per toolset)
- Live chat panel per agent
- Fleet management with bulk config operations
Future adds:
- Memory integration (from any provider incl. custom)
- Skills registry
- Deployment to more baremetal providers (AWS, Azure, GCP, Hetzner etc.)
- Addition of more agent harnesses (e.g. OpenClaw etc)
See it here: https://corellia-frontend.vercel.app/sign-in
Login details: test@user.com + password: Test1234!
README excerpt
<div align="center">
<img src="docs/assets/logo.png" alt="Corellia" width="320" />
<h1>Corellia</h1>
<p><b>A vendor-neutral control plane for AI agents.</b><br/>
Spawn, deploy, govern, and manage agents across any model, any provider, any framework.</p>
<p>
<a href="docs/vision.md">Vision</a> ·
<a href="docs/blueprint.md">Blueprint</a> ·
<a href="docs/stack.md">Stack</a> ·
<a href="docs/changelog.md">Changelog</a>
</p>
</div>
<!--
DEMO VIDEO
Run `pnpm install && pnpm frontend:dev` then `pnpm capture:video`.
Drag docs/assets/login-animation.mp4 into any GitHub issue/PR comment box
to get a CDN URL, then replace VIDEO_URL below and uncomment the block.
<div align="center">
<video src="VIDEO_URL" autoplay loop muted playsinline width="900" />
</div>
-->
---
## What it does
Click **New Agent**. Pick **Hermes**. Name it, choose a model, set a region. Hit deploy.
Thirty seconds later you have a live AI agent — isolated in its own Fly.io microVM, secrets-scoped, health-checked, and fully managed by Corellia. Click the agent's name and start chatting with it directly from the UI.
That's the v1 golden path. Here's what's behind it:
- **Spawn wizard** — "RPG character creation" for agents: harness → name → provider / model → deployment config → deploy. Streams Fly logs in real time as the machine boots.
- **Fleet view** — gallery and list, live status, per-agent actions (start, stop, destroy), bulk ops with a dry-run preview.
- **Fleet i