I don’t even have a clever name for this, but the app I’ve built to manage my personal agent fleet has so much in it at this point. It’s such a strange project because it is inherently _just_ for me, so coupled to my own infra, but works really, really well. It’s built on top of the Claude harness and takes advantage of their channels feature extensively, but is model-agnostic as the harness can run local models too.
The main problem this solved for me was closing the agent->ci pipeline gap. GitHub is deeply embedded via an AWS lambda tailscale proxy:
- I can see issues and PRs in both the global and conversation context.
- Conversations are project and long-lived with stable session identifiers
- CI webhooks arrive immediately and can be acted upon.
- Issues are the canonical control surface
- Agent harness Tasks are used to surface internal planning and status
iPad and macOS layouts share a lot of code too, although the macOS version has an AppKit wrapper to get the native Liquid Glass toolbar looking correct.
I don’t think I’ll ever release the app, but a lot of the pieces are in flight to release:
- tailscale-webhook-proxy sits in the pubic and acts as the verifier and gateway for external messages
- tailnet-webhook-receiver acts as a gateway timeout webhook events to different internal servers
- TartRunnerKit is a receiver plugin that helps orchestrate Tart GitHub runners for specialized workflows
It’s nuts. https://docs.happitec.com/github-aws-tailnet-proxy/
On iOS 27, known schemas prevent me from being able to directly integrate with Siri UI, but I can still have it trigger the shortcut and return the results.
In summary:
- my multi-agent orchestrator app has App Intents
- Shortcuts fetches current active agents and their recent messages, sends them to the cloud model
- Cloud model summarizes, and forms a narrative that knows that agents can collaborate
- Summary is surfaced by Siri UI.
erin sparling