Orbit
A personal-agent iPhone app for the AI you run on your own servers.
Orbit is a native iOS app that gives self-hosted AI agents a calm, polished home. It connects to the models your team already runs — Ollama, vLLM, LM Studio, Hermes, OpenRouter or Claude — so conversations, briefings and plans never pass through a third-party cloud.



- 6+
- agent backends supported
- 0
- third-party services or SDKs
- 100%
- data on the phone & your server
- iOS 18+
- native Swift & SwiftUI
What it is
Teams are running capable open models on their own hardware, but the only way to talk to them is usually a browser tab or a terminal. Orbit turns that server into a personal agent people actually want to use every day: a named assistant with a face, a memory and a home screen of useful things to do.
The app is only a front end. It speaks the OpenAI chat format and the Claude API over HTTPS or your private network, keeps chats on the device, and stores API keys in the iOS Keychain. Switch between agents — the Ollama box at the office, a Hermes agent, Claude — with one tap.
What it can do
Streaming chat
Replies appear token by token, with formatted lists, code blocks with copy, quick-reply options and a stop button.
One-tap agent switching
Keep several servers configured and jump between a local model, a reasoning model and Claude from the chat header.
Briefings
Short posts the agent writes for you from a prompt you control — server status, news on your interests, daily focus.
Playbooks
A catalogue of ready-made requests — health checks, weekly plans, docker-compose files — that drop into the composer.
Missions
Pick a category, shape a goal with the agent in chat, then track its status and progress notes.
Vault
Save useful replies as documents and images as media, then reopen, edit and share them later.
Editable persona
SOUL, IDENTITY and MEMORY files define the agent's personality and context and become its system prompt.
Photos, files & voice
Send photos to vision models, attach text, code or PDF files, and dictate messages on-device.
The iPhone app
Chat, briefings, playbooks, missions and the agent's profile, connected to qwen3:8b running on Ollama.
iOS app








Private by design
Orbit talks only to the agents you configure. There is no Orbit cloud in between.
Data stays on the device
Chats, briefings, missions and the vault live in an on-device database. Nothing is synced to a vendor cloud.
Keys in the Keychain
API keys for OpenRouter, Claude or a private gateway are stored in the iOS Keychain, never in plain preferences.
Straight to your server
Messages go only to the agent endpoint you configure — on your LAN, over a VPN such as Tailscale, or over HTTPS.
No analytics, no SDKs
100% Swift and SwiftUI with zero third-party dependencies: no trackers, ad SDKs or telemetry.
Live health checks
The app pings the active agent and shows online status, latency and available models before you send anything.
Your models, your rules
Choose which model handles which conversation and keep sensitive work on hardware you control.
How it connects
Clients
Your agents
OpenAI-compatible /v1/chat/completions · SSE
Ollama, vLLM, LM Studio, llama.cpp, Hermes, LiteLLM, OpenRouter.
Anthropic API Messages API · SSE
Claude models with your own API key.
Agent gateway (optional) HTTPS · VPN
Your own service that adds tools, approvals and schedules.
Data
On-device store
Chats, briefings, missions, vault and activity (SwiftData).
Keychain & persona files
API keys, plus SOUL / IDENTITY / MEMORY Markdown files.
- The app streams replies over Server-Sent Events from whichever agent is active.
- The persona files are sent as the system prompt, so every backend gets the same personality and memory.
- Reasoning output is filtered so the user sees a clean answer and a "thinking" status.
- Reach a home or office server on the local network, or from anywhere through Tailscale or WireGuard.
Built with
iOS app
Networking
Device
Tooling
From a home lab to your company
Orbit works with the AI you already run, from a single Mac to a shared GPU server or a private gateway.
Individuals & home labs
Point the app at Ollama or LM Studio on a Mac or gaming PC and start chatting in minutes.
- Any OpenAI-compatible server
- Local network or Tailscale
- 8B models run on a 16 GB Mac
Teams with a shared model server
One GPU box or a LiteLLM gateway serves the whole team, each person with their own agent on their phone.
- vLLM / LiteLLM / OpenRouter
- Per-user API keys
- Optional gateway for tools & approvals
Companies & white-label
Your brand, your agents and your App Store listing, connected to your private AI infrastructure.
- Custom branding & agent persona
- App Store / TestFlight / MDM distribution
- Integrations by Singularity Box
Requirements
- Device
- iPhone on iOS 18 or later
- Agent server
- Any OpenAI-compatible endpoint, or a Claude API key
- Network
- Same Wi-Fi, a VPN (Tailscale / WireGuard) or HTTPS
- Local models
- e.g. Ollama with Qwen 3 or Hermes 3 (8B) on 16 GB RAM
- Build
- Mac with Xcode 27 and an Apple Developer account
Built for people who run their own AI
Companies running private LLMs
Give employees a friendly mobile front end for the models you already host, without sending data to a consumer app.
Self-hosters & home-lab users
A real app for the Ollama or LM Studio box at home, with memory, briefings and plans.
AI product builders
A white-label iOS client to put your own agent in front of customers under your brand.
Privacy-conscious professionals
Legal, health and finance work where conversations should stay on hardware you control.
Want Orbit for your organization?
The source code lives in a private repository. Live demos, TestFlight builds and technical walkthroughs are available on request.