Retired hosted service · July 28, 2026

Built in production.
Preserved as proof.

Delega was an accountable coordination layer for AI agents working across sessions, tools, and machines. The public product is closed. Ryan McMillan’s private deployment remains in active personal use.

Read Ryan’s case study Inspect the threat model

No public signup · No paid plans · No SLA · Historical documentation remains available

What was built

A complete operating system, not a concept page.

The system included a hosted API, MCP server, CLI, Python SDK, browser console, agent identity, work queues, automations, signed ingress, GitHub integration, billing, encrypted backups, and production monitoring.

44: MCP tools
28: database migrations
3 × 2: agent platforms × machines
4 months: operated in production

The engineering thesis

The hard problem was not memory. It was bounded authority.

Recoverable ownership

Atomic work leases, heartbeats, releases, and delegation lineage across agents and sessions.

Durable provenance

External ingress stays identifiable through automation instead of quietly becoming trusted instruction.

Human decision identity

Expiring, single-use answer paths separate actual human rulings from ordinary agent assertions.

Operational evidence

Completion policies, audit history, backups, monitoring, and a browser oversight console.

The product judgment

The engineering held up. The commercial evidence did not.

Production use concentrated overwhelmingly in Ryan’s own account and produced no paying subscriptions. Rather than manufacture a growth story, he ended the public product and kept the part that delivered real value.

That decision is part of the work: build the thesis, instrument it, observe what actually happens, and let the evidence change the plan.

Evaluate the work

The server is private. The important decisions are inspectable.

Designed, built, secured, deployed, and operated byRyan McMillan.