Paracoding OSS v2

Build your own agentic fleet.

A chat box forgets, can't act, and can't be trusted with your cloud. Paracoding is the open platform for standing up your own AI fleet on Google Cloud — your own MCP server, a team of role-based agents, durable memory, and a passkey-guarded human gate. AI proposes. You commit.

1connector, many tools
Nrole-based agents
Face IDon the commit key
0unattended paths to root
The idea

A console is one agent in a chat window.
A cockpit is a fleet you strap into.

We got tired of renting our own leverage. So we built the machine we'd actually run a company on — and then we did, on Seaside Mushrooms. This is that machine, open-sourced.

Console

The chat box you rent

  • Forgets everything the moment the turn ends
  • Can talk, but can't safely act on your cloud
  • One prompt trying to be everything
  • No audit trail, no gate, no teammates
  • Stops the second you close the tab
Cockpit

The fleet you own

  • Durable per-role memory that survives every turn
  • Acts on real infrastructure — behind your Face ID
  • Specialized agents with roles, not one mega-prompt
  • A shared bus, a work board, and a human gate
  • A real computer that turns on, works, turns off
What's in the box

Six parts, one machine

Every piece is deployed to your Google Cloud project. Nothing phones home; your fleet is yours.

🔌

Your own MCP server

A serverless control plane on Cloud Run — one connector exposing role-aware tools. One-command deploy from the scaffold. The smallest server that does real work.

Cloud Run · MCP · OAuth 2.1
🧑‍🚀

A fleet of roles

Advisor, Infra, Publisher, Security, Ads, Ghost — each an agent with its own private memory and a shared drop zone. Roles beat one giant super-prompt, every time.

Per-role identity · private + shared memory
🔀

An event bus

Agents hand each other work with no human relay: work items, agent-to-agent messages, wake events. One agent files a job; another picks it up and answers.

ask_agent · work board · wake events
🔐

The human gate

The most important design decision: AI can propose anything, but nothing runs until a human commits it — with a passkey. Face ID for your infrastructure.

WebAuthn · staged jobs · batch approve
🛩️

A real cockpit

A persistent cloud workstation that boots on demand, does the heavy work, and shuts off — with a full Linux desktop streamed right into your browser or any RDP client.

Compute Engine · streamed desktop
🍄

Dog-fooded, not demo-ware

We refused to ship anything we wouldn't run our own business on. The fleet runs Seaside Mushrooms — site, content, ads, ops — so the platform is proven on real work.

Proven in production
The cockpit

The jet, assembled

Cockpit + chat + fleet + gate + workstation as one machine you strap into.

You + Passkey the commit key Human Gate AI proposes · you commit MCP Control Plane role-aware tools Agent Fleet roles + bus Gated Executor + cockpit workstation Data lake · durable memory (Firestore + GCS)

Everything an agent needs to actually do the job

  • Chat console — talk to the fleet in plain language; bring your own Claude or Gemini key.
  • One-tap approvals — the gate shows every staged job; approve them in a batch with a single Face ID.
  • The 12-hour ops window — authenticate once each morning, and approved work runs as you until you log out.
  • A streamed desktop — a persistent Linux workstation with Chrome and tools, reachable from the cockpit or any RDP client.
  • Auto-expiring jobs — stale proposals clean themselves up, so the queue is always the truth.
Security by design

Giving AI your cloud shouldn't mean giving up control.

The whole platform is built on one rule: there is no unattended path to root. Your agents can draft, plan, and stage anything — but the only way a command touches your infrastructure is your passkey, on your device, in the moment. Least-privilege identities, secrets that never live in an image, and short-lived credentials mean a leak is never game over.

Read the full security model →
  • Passkey commit — every approval is WebAuthn / Face ID
  • Runs as you — short-lived Google identity, revoked on logout
  • Least privilege — service accounts scoped to the job, never owner
  • Secrets stay put — Secret Manager, never baked into images
  • SHA-pinned deploys — verified, additive, restore-on-fail
Open-core

Free and open at the core

Clone it, deploy it to your own GCP, and sit in your own cockpit. Premium unlocks fund development and keep the lights on — the core is yours to keep.

Free & open

The core fleet

$0
  • Serverless MCP server, one-command deploy
  • Two-agent example fleet (Advisor + Infra)
  • Passkey human-gate for applying jobs
  • Bring your own Claude / Gemini key
  • Runs entirely in your own Google Cloud
★ Get it on GitHub soon
Premium · coming soon

Cockpit unlocks

$99 / feature
  • Dashboard & live chat widget
  • One-tap Windows-VM deploy
  • Shielded RDP button
  • 1-hour free trial key per feature
  • Lifetime entitlement — unlimited keys
Buy — coming soon
Bundle · coming soon

The 10-agent pack

$99
  • All ten production agents
  • Standing orders + boot configs
  • The exact roles that run Paracoding
  • Drop-in on your own control plane
Buy — coming soon

Sign in with Google, spin up a free fleet, and support the work by buying a key — or by grabbing some mushrooms from Seaside. Pricing and store finalize at launch.

Ships on GitHub

Strap in. The repo is on final approach.

We're extracting the turnkey release now — public cloud URLs, no domain required, cockpit ready on first boot. Star the repo to get the launch, or read exactly how it's secured while you wait.

★ github.com/paracoding-ai publishing soon Read the security model →
Paracoding OSS is a platform for running AI agents against real infrastructure. It is designed so that a human, holding a hardware passkey, is the only party who can authorize a privileged action — but any system that grants automation access to cloud resources carries risk. Deploy it in a project you control, review what your agents stage before you approve it, and keep least-privilege and billing alerts switched on. Premium features, pricing, and availability are described here for planning and may change before launch.