Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Mobile & Edge LLMs

Phantasm Review: Self-Hosted AI Chat App for iPhone

·6 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Phantasm (version 1.2 on the App Store, released on 25 July 2026) is a free iPhone chat app that talks to an OpenAI-compatible AI server you run yourself, so the models run on your hardware, not on the phone. The developer says it has no accounts, no analytics and no telemetry, and every statement here comes from the App Store listing, the website, the privacy policy and the GitHub repository; PromptQuorum has not tested it.

Key Takeaways

  • Maintenance: App Store version 1.2 is dated 25 July 2026, and the last repository commit, a build bump to 1.3, is dated 16 August 2026; the App Store did not list 1.3 on 13 November 2026.
  • Cost: free download; the repository README calls accounts, billing and cloud sync a deferred premium layer.
  • Platform: iPhone with iOS 18 or later; the listing also shows Mac and Apple Vision requirements but says the app is designed for iPhone.
  • Source: the website and privacy policy call the app and server open source, but the repository showed no licence file when checked.
  • Maker: Benjamin Carman is the App Store seller, and the first release was on 8 July 2026.

📍 In One Sentence

Phantasm is an iPhone chat client for Ollama, vLLM, llama.cpp and other OpenAI-compatible servers that you host yourself.

💬 In Plain Terms

You enter the address and token of the computer that runs your models, and the app then chats with it, attaches photos and files and, with the optional server, searches the web.

📌Note: This review is based on the App Store listing, the website, the privacy policy, the support page and the GitHub repository, checked on 13 November 2026. PromptQuorum has not installed, run or benchmarked the app or the orchestrator, so speed, answer quality and setup effort are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Phantasm is a free iPhone chat app, listed under the developer Benjamin Carman, that connects to an AI backend you run yourself, such as Ollama, vLLM or llama.cpp, and needs no account. It does not run models on the phone: the app is the client, and an optional companion server adds web search and image generation. This review is based on the App Store listing, the website, the privacy policy and the GitHub repository, checked on 13 November 2026. PromptQuorum has not installed or run it.

Phantasm Review: Self-Hosted AI Chat App for iPhone

What Is Phantasm?

Phantasm is an iOS chat client for AI models that you host yourself, not an app that contains a model. The App Store lists it as "Phantasm" from the seller Benjamin Carman, it first appeared on 8 July 2026, and the website calls it a "fast, self-hostable AI chat app for iOS that talks to a backend you control".

The project has two parts: the SwiftUI app, which takes a server URL and a token, and an optional Rust server called the orchestrator that runs next to Ollama and ComfyUI and adds web search, image generation and a tool loop. Pointed at a bare Ollama instance, the app falls back to plain chat, according to the README.

Get It

Phantasm is a free App Store download for iPhone; the sources name no Android, Windows or web version.

Option
How to get it
iPhonePhantasm on the App Store, free download
Websitegetphantasm.com
Privacy policygetphantasm.com/privacy.html
Backend serverOrchestrator on GitHub, optional, starts with docker compose
Source codegithub.com/benthecarman/phantasm, no licence file

Version reviewed: 1.2, released on 25 July 2026 per the App Store; the app is 19.1 MB. The app has no built-in model, so you also need a server to chat with. The listing shows the price as free and names no in-app purchases.

Requirements

The App Store lists iOS 18.0 as the minimum, and the app cannot chat without a server that you provide.

Operating system

What the sources say:
iOS 18.0 or later on iPhone

Backend

What the sources say:
An OpenAI-compatible server you provide, such as Ollama, vLLM or llama.cpp, reachable from the phone

Connection

What the sources say:
A server URL and a token; the token is kept in the iOS Keychain

Orchestrator (optional)

What the sources say:
Docker Compose; adds Brave web search, ComfyUI image generation and a tool loop

Account

What the sources say:
None; the developer says there are no accounts

The sources name no minimum iPhone model or memory; the hardware you need depends on the models your own server runs. Reaching the server from outside your home network is your own job: the README says you use a server URL that is already reachable from your device.

How to Get Started

The steps follow the developer's own description of the app; PromptQuorum has not run them.

  1. 1
    Run a backend
    Why it matters: Start Ollama or another OpenAI-compatible server on a machine you own; optionally add the orchestrator with docker compose for web search and image generation.
  2. 2
    Install the app
    Why it matters: Get Phantasm from the App Store; no account is needed.
  3. 3
    Enter the URL and token
    Why it matters: Add the server address and token in Settings; the app finds the available models and capabilities on its own.
  4. 4
    Chat
    Why it matters: Pick a model, attach photos or files, and switch web access and image generation on or off for each chat.

Features Confirmed by the Sources

Every item below comes from the App Store description, the website or the repository; none has been independently tested.

  • Flexible backends. Any OpenAI-compatible endpoint, with several saved profiles for home network, VPN or tunnel setups and capability detection at runtime.
  • Chat display. Streaming answers with markdown, syntax-highlighted code with a copy button, charts, inline math and the model's reasoning when a thinking model is selected; version 1.2 adds token and speed estimates.
  • Media. Photo and file attachments for vision models, inline generated images, and audio and video generation with compatible servers since version 1.2.
  • Voice and system integration. On-device dictation, read-aloud replies, Siri and Shortcuts, and long generations that pick up again after the app was in the background.
  • Tools. Per-chat switches for web access and image generation, plus optional location, calendar and Apple Health tools that are off by default and shared only with your own backend.

Data Flow and Privacy

The privacy policy, last updated on 2 July 2026, says Phantasm has no accounts, no analytics, no telemetry and no servers of the developer in the middle. Messages go only to the backend URL you enter, conversations are stored on the phone, and speech transcription runs on the device. The App Store privacy label reads "Data Not Collected".

Where your data travels depends on your backend. The website says conversations never leave your network, but the app also supports VPN and tunnel profiles, so a remote setup sends traffic over whatever route you build. The optional device tools share location, calendar or Health data with your backend only when you switch them on for a chat. The website is served through Cloudflare, which processes standard request data, and the policy says it sets no cookies.

📌Note: These are the developer's declarations, not audit results, and PromptQuorum has not inspected the app's network traffic. When the orchestrator's web search is on, the server queries Brave Search according to the README, so those queries leave your network.

Who Should Use It

  • Ollama, vLLM or llama.cpp owners who want a native iPhone front end for models already running at home.
  • Privacy-minded users who want an app with no account and no analytics, as the developer describes it.
  • Home-lab tinkerers who are comfortable running Docker Compose for web search and image generation.

What We Could Not Verify

  • Performance. Speed, battery use and behaviour in very long chats are untested.
  • Orchestrator. Web search, ComfyUI image generation and the tool loop were not run; the README calls the project an MVP build.
  • Licence. The site calls the app and server open source, but the repository showed no licence file on 13 November 2026, so the terms for reuse are not stated.
  • Other devices. The listing shows Mac and Apple Vision requirements yet says the app is designed for iPhone; iPad and Mac support is not described.
  • Latest build. The repository holds a version 1.3 build bump from 16 August 2026 that the App Store listing did not show on 13 November 2026.

Competitors and Alternatives

LM Mini

Platforms:
iPhone, iPad, Mac listing
Price / license:
Free + in-app / closed source
Key difference:
Also offers remote access and optional cloud APIs
Articles about LM Mini (1)

Also mentioned in:

Reins

Platforms:
iPhone, iPad, Mac listing
Price / license:
Free + Pro / closed source (v2)
Key difference:
Connects to your server and can also run MLX models on the device
Articles about Reins (3)

Confirm each app's current price, license and platforms on its own listing before choosing.

Compare in directory

Frequently Asked Questions

Does Phantasm work without a server?

No. The App Store listing says it requires an OpenAI-compatible server that you provide, for example Ollama on your own machine; the privacy policy says only speech transcription runs on the phone.

Can I use Phantasm away from home?

The app takes any reachable URL and the listing mentions VPN and tunnel profiles, but the sources describe no built-in remote-access service, and the README lists a managed relay as deferred.

Which models does Phantasm support?

Any model your backend serves, because the app bundles none. Vision input, reasoning display and image, audio or video generation depend on what your backend offers.

Can I use Phantasm with Siri?

The listing says you can ask Phantasm from Siri and Shortcuts; how deep that integration goes is not described.

Verdict

Phantasm is a young project with recent commits that gives an iPhone a free, account-free way to reach a home Ollama, vLLM or llama.cpp server, and its privacy policy declares no data collection. Against that, it is only a client, it needs a server you run, its repository showed no licence file on 13 November 2026 although the site calls it open source, and its newest repository build was not on the App Store when checked. It suits people who already host models and want a native front end; anyone who needs audited privacy claims should treat the policy as a declaration, not proof.

Sources

← Back to Power Local LLM

MCP