juli-os iOS — the remote for your agent OS

The native iOS surface of the juli-os ladder: a pocket remote for a self-hosted agent platform. Your Mac runs the engine; your phone runs this app.

Screenshots below were taken against a real engine instance seeded with sample data (Acme Labs / Northwind) — every screen you see is the live app, not a mockup.

The tour

Order by talking. The chat intake asks clarifying questions, then hands you a work-order card — nothing is created until you confirm. Voice mode uses local VAD so speech is only streamed while you're actually talking.

Flow is your queue. Pending human approvals sit on top (waiting_human), then running and finished jobs — each row shows the routing target and state.

Every job is a casefile. Open a job and you get the full paper trail: the route chain (why this email landed in this agent), the original message, the agent's understanding, and the draft reply with its deliverables.

Gates are the point. Nothing leaves without you. The send gate shows the draft, its attachments, and pre-send checks (recipient matches thread, thread intact, attachment hash verified) — approve or reject; the decision is written to the ledger.

Agents as a map. The Agents tab renders the live multi-agent topology from the engine's declarations — companies, domains, working state — plus recent dispatch decisions.

Artifacts & settings. Deliverables are content-addressed files with roles (report / body / share); Settings holds the engine address (stored in the Keychain), voice credentials, and VAD tuning.

What you get

• Chat intake — conversational work orders: describe a task, confirm, and it becomes a tracked job
• Voice control — speech-to-text with local VAD; bring your own Azure Speech key in Settings
• Gate workbench — human-approval queues with content-hash binding: approve or reject what the engine wants to send
• Routing map & agents — live multi-agent topology, session states, recent dispatch decisions
• Terminals & artifacts — read-only agent terminal snapshots, artifact previews (HTML/report/body roles)

Requirements

• Xcode 16+, xcodegen, CocoaPods
• A running juli-os engine (see @juli-os/ledger / @juli-os/artifacts / @juli-os/routing — or the full stack)

Build

cd Makro
xcodegen generate
pod install
open Makro.xcworkspace
# pick the Makro scheme, a simulator, run — no signing needed for the simulator

The app stores your engine address and credentials in the Keychain; set the server URL in Settings on first launch.

Reproducing the screenshots

The shots above come from a plain engine instance seeded with sample data (fake companies acme / northwind, two jobs parked at their send gates). Run the engine with any config, point the app at it, and you get the same surfaces with your own data.

Architecture notes

• SwiftUI app, no third-party app-level dependencies (only the Azure Speech Pod for voice)
• Talks to the engine over its HTTP/SSE/WebSocket API (APIClient*)
• The project file is generated from project.yml — regenerate after edits, never hand-edit project.pbxproj

License

Apache-2.0