# Delega — Historical Technical Archive

Hosted access was retired on July 28, 2026. The Cloudflare deployment remains in active personal use by Ryan McMillan and is not available for public signup, agent onboarding, billing, or general API access.

## Canonical project record

- Portfolio case study: https://ryanmcmillan.com/delega
- Architecture and threat model: https://delega.dev/architecture
- Historical/as-built documentation: https://delega.dev/docs
- Changelog: https://delega.dev/changelog
- Public source: https://github.com/delega-dev
- Historical OpenAPI contract: https://api.delega.dev/v1/openapi.json
- Health: https://api.delega.dev/health

## Retirement contract

- Existing owner credentials may continue to use `api.delega.dev`.
- Authenticated accounts outside the owner allowlist receive `403 service_retired`.
- Public signup, verification, claim, recovery, and billing entry points receive `410 hosted_service_retired`.
- There are no public plans, public support commitment, or public onboarding path.
- Public client repositories and packages are retained as engineering artifacts. Their presence does not imply hosted-service availability.

## System summary

Delega was a production coordination system for AI agents, built around leased ownership, delegation lineage, durable provenance, bounded automation, and explicit human decision identity. The public hosted service retired on July 28, 2026. Ryan McMillan continues to use a private deployment, while the architecture and public clients remain available as portfolio evidence.

Delega was implemented as a production coordination system for AI agents across sessions, machines, and runtimes. Its surfaces include:

- a Cloudflare Workers REST API and D1 data model;
- an MCP server used by Codex, Claude Code, and Hermes;
- a Node.js CLI and Python SDK;
- leased task claims, heartbeats, handoffs, and fleet-attention triage;
- delegation chains and named agent roles;
- durable task context with provenance and optimistic concurrency;
- cross-task decision recall;
- recurring tasks and bounded in-process automation;
- signed inbound connectors and HMAC webhooks;
- signed Decision Answer links for narrow human rulings;
- evidence-required completion policies;
- a private browser task console.

## Trust model

The architecture favors explicit identity, least-privilege keys, server-enforced account boundaries, fail-closed configuration, hashed API keys, signed external inputs, redacted telemetry, and auditable task history. The public architecture page distinguishes implemented controls from known limitations.

## Historical documentation

The documentation and changelog describe the system as built. Examples that refer to package installation or API operations are retained for technical verification only. They are not an invitation to create an account or access the private deployment.

Designed, built, secured, deployed, and operated by Ryan McMillan.