TALARIA

A native iOS client for your self-hosted Hermes AI agent.

SELF-HOSTED  ·  SWIFT NATIVE  ·  DIRECT API  ·  SENSORS + VOICE

iOS 26+ Swift 6 SwiftUI Hermes Agent Tailscale Build from source
VIEW ON GITHUB SETUP GUIDE

NOTE  —  Talaria is an independent community project. It is not affiliated with, endorsed by, or part of Nous Research or the official Hermes Agent project.

// INTERFACE

Every screen, purpose-built.

From initial pairing through streaming chat, voice, and diagnostics — Talaria's UI is designed around the Hermes agent experience.

// MAIN FLOWS

Talaria onboarding / pairing screen HANDSHAKE
Talaria chat screen CHAT
Talaria voice mode VOICE
Talaria directives screen DIRECTIVES
Talaria settings overview SETTINGS

// SETTINGS SCREENS

System settings SYSTEM
Uplink / connection settings UPLINK
Model selection MODELS
Voice settings VOICE
Appearance / HUD settings APPEARANCE
Sessions management SESSIONS
Diagnostics screen DIAGNOSTICS

// WHAT IT IS

Your phone. Your agent. Your backend.

Talaria is a native SwiftUI iPhone app that connects directly to a Hermes AI agent you host yourself. There is no cloud, no managed relay, no subscription — just your device talking to your machine over your network.


Chat talks directly to the Hermes Gateway API. Sensors flow through a lightweight sidecar relay. Model switching is handled by a local shim. Everything is auditable, replaceable, and yours.

Talaria chat interface

// ARCHITECTURE

How it connects.

Three independent paths — chat, sensors, and model control — each talking to a dedicated service on your host machine.

iPhone (Talaria)
  │
  ├─ Chat & sessions  ──────→  Hermes Gateway      :8642
  │    SSE streaming, sync                 hermes gateway run
  │    Bearer auth                         Sessions API + all platforms
  │
  ├─ Sensor data  ──────────→  HermesMobile Relay  :8000
  │    Location, HealthKit, motion         sidecar process (Python/uvicorn)
  │    Background upload                   → hermes_mobile MCP tools
  │
  └─ Model switching  ──────→  Models Shim         :8765
       Live model list + swap              tools/models-shim/shim.py
       Per-session, no restart             (optional)
:8642
HERMES GATEWAY
The Hermes Sessions API. Chat, streaming SSE, and all enabled platforms in one process. Run with hermes gateway run.
:8000
HERMES MOBILE RELAY
Lightweight Python sidecar that receives sensor uploads from the phone and delivers them to Hermes via hermes_mobile MCP tools.
:8765
MODELS SHIM
Optional. Exposes the full live model list from your Hermes install and allows per-session model switching without a gateway restart.
TAILSCALE
NETWORK LAYER
Recommended for remote access. The app uses Tailscale hostnames and IPs to reach all three services without exposing them to the public internet.

// CAPABILITIES

What Talaria does.

💬
STREAMING CHAT
Server-sent events from the Hermes Sessions API. Markdown, code blocks, inline images, and agent file downloads — all rendered natively.
🎙️
VOICE MODE
Real-time WebRTC speech-to-speech via the relay. Server-side voice, continuous mic, mute/barge-in, and multimodal image support.
📡
SENSOR PIPELINE
Location, 11 HealthKit metrics, and CoreMotion activity upload in the background. Your agent gets live context about you — and you own all the data.
🔀
LIVE MODEL SWITCHING
Pick from your full provider roster mid-session. Switches take effect on the next session without restarting any service.
📁
AGENT FILES
Files your agent generates are surfaced as tappable share bubbles in chat. Download, save, or forward directly from the conversation.
⚙️
FULL SETTINGS SUITE
System, Uplink, Models, Voice, Appearance, Sessions, Diagnostics — everything configurable in-app, no config file edits needed.
🔍
DIAGNOSTICS
Live connection health, port status, session counts, and service state — all visible from the app without touching the host terminal.
🔒
DIRECT, NO CLOUD
Chat goes phone → Hermes Gateway. Nothing passes through a third-party server. Bearer token auth. You control the keys.

// GETTING STARTED

Self-host in five steps.

You need a machine running Hermes, a Tailscale network, and Xcode to build the app. There is no TestFlight or App Store distribution.

01
INSTALL AND RUN HERMES AGENT

Follow the Hermes Agent install instructions for your host OS (macOS or Windows). Verify hermes is in your PATH and a profile is configured.

02
START THE HERMES GATEWAY

Run hermes gateway run on your host. This starts the Sessions API on :8642. On Windows, use NSSM or a Scheduled Task for persistence. On macOS, use a launchd agent. Ensure the port is reachable from your Tailscale IP.

03
DEPLOY THE RELAY SIDECAR

Clone this repo and run the HermesMobile relay from relay/. Set AGENT_FILES_DIR in your .env for agent file downloads. The relay handles sensor delivery and the optional voice WebRTC bootstrap. Bind to 0.0.0.0 so your phone can reach it over Tailscale.

04
BUILD TALARIA IN XCODE

Open the Xcode project (requires Xcode for iOS 26 SDK). Select your physical iPhone as the run destination. Build and run. You will need an Apple Developer account to sign the app for your device.

05
PAIR VIA THE ONBOARDING SCREEN

On first launch, Talaria shows the pairing screen. Enter your host's Tailscale IP or hostname, the gateway port (8642), and your API_SERVER_KEY. The app connects directly — no account, no cloud login.

⚠️ If you use iCloud Private Relay, disable it for Tailscale IPs or the app will not be able to reach your services.

// TECH STACK

Built on.

Swift 6 SwiftUI iOS 26+ Xcode (iOS 26 SDK) XcodeGen Hermes Agent Sessions API (SSE) WebRTC HealthKit CoreLocation CoreMotion Python · uvicorn Tailscale NSSM (Windows) launchd (macOS)