Key Takeaways
- Local mode: Liquid AI's LFM2 models (350M/700M/1.2B/2.6B parameters) run on-device via LEAP — fully offline once downloaded, no account needed
- Cloud mode: OpenRouter integration reaches models like Meta Llama 3 and OpenAI GPT-4 — requires your own OpenRouter API key and an internet connection
- Custom-backend mode: connects to a self-hosted LM Studio or Ollama server on your own network — requires that server to be running and reachable
- Free app, no Liquid AI subscription; OpenRouter usage is billed separately by OpenRouter/the model provider, not by Liquid AI
- Version 2.2.4 on the App Store, 4.5 out of 5 stars from 634 ratings, 13+ age rating
- Platforms: iPhone (iOS 18+), iPad (iPadOS 18+), Mac (Apple Silicon, macOS 15+), Apple Vision (visionOS 2+), and Android via Google Play
- Developer: Liquid AI, Inc., which also builds the LEAP SDK and the LFM2 model family used in the app's local mode
📍 In One Sentence
Liquid Apollo is a free app for iPhone, iPad, Mac, Apple Vision, and Android that runs Liquid AI's own LFM2 models fully on-device via its LEAP technology, but also — as a first-class, not hidden, feature — connects to cloud models through OpenRouter and to self-hosted LM Studio or Ollama backends, so only its local mode is actually offline.
💬 In Plain Terms
Think of Liquid Apollo as three apps in one: an offline chat app that talks to small AI models stored on your own phone, a remote control for cloud AI models that reads your own OpenRouter account, and a remote control for an AI server running on your own computer at home. Only the first mode works with no internet at all.
📌Note: This review is built on Liquid Apollo's own App Store and Google Play listings, Liquid AI's own liquid.ai site and press materials, and the LFM2 model cards Liquid AI has published. PromptQuorum has not independently benchmarked Liquid Apollo's on-device inference speed, nor audited network traffic to confirm no data leaves the device in local mode — those are the developer's stated design, not third-party-verified guarantees.
What Is Liquid Apollo?
Liquid Apollo is a chat client built by Liquid AI, Inc. that lets you talk to AI models in three different ways: a model running on your own device, a hosted model reached over the internet through OpenRouter, or a model running on a server you host yourself. Liquid AI describes it, in the app's own listing, as "your own customizable client for accessing language models from all around the web" — cloud access is presented as a core feature, not an afterthought.
- Local models: powered by LEAP, Liquid AI's own on-device inference technology, runs Liquid AI's LFM2 model family
- OpenRouter support: bring your own OpenRouter API key to reach models such as Meta Llama 3 and OpenAI GPT-4
- Custom backend support: connect to a self-hosted LM Studio or Ollama instance on your own computer
- Developer mode (added in version 2.2.4): shows per-message token counts, timings, and tokens-per-second, for both local and remote models
- Free download, no account required to use local mode
Get Liquid Apollo
Liquid Apollo is free to download from both the Apple App Store and Google Play — there is no purchase required to try any of its three modes.
iPhone / iPad
- Where to get it:
- App Store
- Requirements:
- iOS 18.0+ / iPadOS 18.0+
Mac
- Where to get it:
- App Store
- Requirements:
- macOS 15.0+, Apple Silicon (M1 or later)
Apple Vision
- Where to get it:
- App Store
- Requirements:
- visionOS 2.0+
Android
- Where to get it:
- Google Play
- Requirements:
- Package: ai.liquid.chatapp
For local-only use, the setup stops at downloading an LFM2 model from inside the app. Cloud mode additionally requires creating your own account at OpenRouter and generating an API key; custom-backend mode additionally requires an LM Studio or Ollama server already running and reachable from your phone or Mac.
Local Mode: LEAP and LFM2 Models
Liquid Apollo's on-device mode is powered by LEAP (Liquid Edge-AI Platform), a developer SDK Liquid AI also ships separately for other app builders, and it runs Liquid AI's own LFM2 model family rather than a third-party model. This is the only one of Liquid Apollo's three modes where a conversation can, by design, stay entirely on the device with no network connection.
LFM2 models come in four published sizes: LFM2-350M, LFM2-700M, LFM2-1.2B, and LFM2-2.6B parameters. According to Liquid AI's own model documentation, they are built from "structured, adaptive operators" rather than a standard transformer architecture, trained on roughly 10 trillion tokens, and designed specifically for fast inference on phone-class CPUs, GPUs, and NPUs. Liquid AI states the smallest models in its library run in as little as 300 MB and that the library runs on devices with 4 GB of RAM or more.
- Model family: LFM2-350M, LFM2-700M, LFM2-1.2B, LFM2-2.6B — all published by Liquid AI, not third-party fine-tunes
- Architecture: Liquid AI describes LFM2 as using structured, adaptive operators, not a conventional transformer stack
- Training: Liquid AI states its LFM2 models were trained on approximately 10 trillion tokens
- Footprint: Liquid AI states its smallest library models can run in as little as 300 MB, on devices with 4 GB of RAM or more
- Connectivity: once a model is downloaded, local-mode chat requires no internet connection
Cloud Mode: OpenRouter
Liquid Apollo's OpenRouter mode is a first-class, clearly advertised feature, not a hidden option — the app's own description invites you to "bring an OpenRouter API key to chat with every model on the web." This mode requires an internet connection and an account you create yourself at OpenRouter, a third-party service unaffiliated with Liquid AI that provides API access to many hosted AI models.
Once you paste an OpenRouter API key into Liquid Apollo, the app can reach "nearly every model" available through OpenRouter, per its own listing — named examples include open-source models such as Meta Llama 3 and closed-source models such as OpenAI's GPT-4. Messages sent in this mode leave your device and are processed by whichever model and provider you select through OpenRouter, exactly as they would be in any other OpenRouter-connected client.
- Requires your own OpenRouter account and API key — Liquid AI does not provide or resell OpenRouter access inside the app
- Requires an active internet connection; this mode does not work offline
- Reaches hosted models including open-source options (for example, Meta Llama 3) and closed-source options (for example, OpenAI GPT-4)
- Billing for model usage in this mode goes through OpenRouter and the underlying model provider, not through Liquid AI
- Any conversation sent through this mode is, by design, transmitted off the device to OpenRouter's infrastructure and the selected model provider
Custom Backend: LM Studio and Ollama
Liquid Apollo's third mode connects to a self-hosted LLM server you run yourself, using tools such as LM Studio or Ollama, and the app's own description frames this as building "your own private ChatGPT-like mobile app for your family." Version 2.2.4 specifically added streaming reliability fixes for this mode, alongside the developer-mode token/timing stats that also apply here.
- Requires a separately installed and running LM Studio or Ollama server, typically on a computer on your own network
- The phone or Mac running Liquid Apollo must be able to reach that server over the network — this is not a cloud service Liquid AI operates
- Data handling in this mode depends entirely on how you configure and expose your own LM Studio or Ollama instance, not on anything Liquid Apollo itself controls
- This mode is functionally similar to what apps like Arbiter and AnythingLLM Mobile offer for connecting a phone client to a home-run model server
Pricing
Liquid Apollo itself is free, with no in-app purchases or subscription listed on the App Store or Google Play. The app's cost structure differs sharply by mode.
Local (LFM2 via LEAP)
- What it costs:
- Free
- Who bills you:
- Nobody — no account or payment involved
Cloud (OpenRouter)
- What it costs:
- Pay-per-use, set by OpenRouter and the underlying model provider
- Who bills you:
- OpenRouter, via the API key you supply — not Liquid AI
Custom backend (LM Studio / Ollama)
- What it costs:
- Free (assuming you already run the server); your own hardware and electricity cost applies
- Who bills you:
- Nobody through Liquid Apollo — cost is whatever running your own server costs you
Liquid AI does not charge its own subscription on top of OpenRouter usage — the app itself has no listed in-app purchases. Confirm current OpenRouter pricing on openrouter.ai before relying on a specific model's per-token cost, since provider pricing changes independently of Liquid Apollo.
Platforms
iPhone
- Requirement:
- iOS 18.0 or later
- Note:
- Primary supported platform per the App Store listing
iPad
- Requirement:
- iPadOS 18.0 or later
- Note:
- Same app and model library as iPhone
Mac
- Requirement:
- macOS 15.0 or later, Apple Silicon (M1 or later)
- Note:
- No Intel Mac support listed
Apple Vision
- Requirement:
- visionOS 2.0 or later
- Note:
- Listed as compatible; this review did not hands-on test the Vision Pro experience
Android
- Requirement:
- Google Play, package ai.liquid.chatapp
- Note:
- Android support arrived after the iOS launch; verify current OS-version requirements on the Google Play listing
Liquid Apollo has no Windows or Linux desktop version. For a custom-backend setup running on a Windows or Linux PC, the LM Studio/Ollama server itself can run on those platforms even though the Liquid Apollo client cannot.
Privacy: What Stays On-Device and What Does Not
Liquid Apollo's App Store privacy label states the developer does not collect any data from the app itself, and no account is required to use local mode. That claim covers the app's own data collection — it does not, and cannot, describe what happens to a message once you choose to send it to OpenRouter or a self-hosted backend, since those are separate services outside Liquid AI's control.
The honest way to think about privacy here is per-mode, not per-app: local LFM2 chat has no network path for your message content to leave the device, by architecture, not just by policy. OpenRouter mode sends message content to OpenRouter and the model provider you select, governed by their own privacy terms. Custom-backend mode sends message content to whatever server you configured, governed by how you set that server up.
- No account required for local mode; Liquid AI states the app collects no data from itself
- Local LFM2 chat has no server-side component to send data to — this is an architectural property of on-device inference, not solely a policy statement
- OpenRouter mode necessarily transmits message content off-device to OpenRouter's infrastructure and your selected model provider — review OpenRouter's own privacy policy if this matters to you
- Custom-backend mode transmits message content to your self-hosted server — privacy here depends on your own server configuration, not on Liquid Apollo
- This review has not independently audited Liquid Apollo's network traffic to verify the no-data-collection claim for local mode
Who Built Liquid Apollo?
Liquid Apollo is built by Liquid AI, Inc., the same company that develops the LEAP on-device inference SDK and the LFM2 model family used in the app's local mode. Liquid AI positions Apollo partly as a public showcase of LEAP, letting developers try LFM2 models on their own phone before integrating LEAP into their own apps.
- Developer: Liquid AI, Inc.
- Also builds: LEAP (the on-device inference SDK Liquid Apollo's local mode runs on) and the LFM2 model family
- Distribution: Apple App Store and Google Play
Trade-Offs
Genuine on-device mode
- What it means in real use:
- LFM2 models run fully offline via LEAP once downloaded — no internet needed for local chat.
- Limitation / caveat:
- LFM2 models are small (up to 2.6B parameters) by design for on-device speed, so capability is well below frontier cloud models.
Free app, three access modes in one client
- What it means in real use:
- No purchase needed to try local, cloud, or self-hosted access from a single app.
- Limitation / caveat:
- Only one of the three modes (local) is actually offline — the app's flexibility can make it easy to assume more privacy than a given mode provides.
OpenRouter access to larger models
- What it means in real use:
- Reach models like Meta Llama 3 or OpenAI GPT-4 from the same interface used for local chat.
- Limitation / caveat:
- Requires your own OpenRouter account and API key, and usage is billed by OpenRouter separately from the free app.
Custom backend support
- What it means in real use:
- Turn a home LM Studio or Ollama server into a mobile chat app for yourself or your family.
- Limitation / caveat:
- Requires you to set up and maintain that server yourself; Liquid Apollo does not host or manage it.
Cross-platform (iOS, Android, Mac, Vision)
- What it means in real use:
- Wider platform reach than most Apple-only local-AI apps in this category.
- Limitation / caveat:
- No Windows or Linux desktop client, so a custom-backend server on those OSes needs a different companion app or a web UI.
Liquid Apollo vs. Alternatives
Liquid Apollo
- Local-only or hybrid:
- Hybrid: local LFM2 + OpenRouter cloud + custom backend
- Platforms:
- iPhone/iPad/Mac/Vision/Android
- Price:
- Free (OpenRouter usage billed separately)
- Key difference:
- Only mobile app running Liquid AI's own LFM2 models on-device; cloud mode is a first-class feature, not an edge case
Articles about Liquid Apollo (1)
- Liquid Apollo Review (2026): On-Device LFM2 Chat, Plus Optional CloudUpdated September 12, 2026
Private LLM
- Local-only or hybrid:
- Local-only, no cloud path
- Platforms:
- iPhone/iPad/Mac (Apple only)
- Price:
- $4.99 one-time purchase
- Key difference:
- Best pick if you want zero cloud option at all, with a large curated model library
Enclave AI
- Local-only or hybrid:
- Local-only, no cloud path
- Platforms:
- iPhone/iPad/Mac (Apple only)
- Price:
- Free, with in-app purchases
- Key difference:
- Free local-only alternative with no cloud mode at all, including for voice chat
Layla
- Local-only or hybrid:
- Hybrid: local GGUF models + optional Layla Cloud
- Platforms:
- Android and iOS
- Price:
- $19.99 one-time, plus in-app purchases
- Key difference:
- Roleplay/character focus; cloud mode is opt-in and off by default, unlike Liquid Apollo's OpenRouter mode
Arbiter
- Local-only or hybrid:
- Hybrid: local models + LAN connection to LM Studio/Ollama
- Platforms:
- iOS only
- Price:
- Freemium
- Key difference:
- Closest structural match to Liquid Apollo's custom-backend mode, but no internet-based cloud mode like OpenRouter
Platform, price, and feature details for third-party apps change frequently — verify current specifics on each app's own listing before deciding.
Who Should Use Liquid Apollo
- Developers evaluating LEAP or LFM2 for their own app. Liquid Apollo is a direct, no-code way to try Liquid AI's own on-device models before integrating the LEAP SDK into a separate project.
- Readers who want one app covering local, cloud, and self-hosted access. If you already use OpenRouter or run your own LM Studio/Ollama server, Liquid Apollo can be a single front-end for all three instead of juggling separate apps.
- Android and iOS users who want the same client on both platforms. Unlike most Apple-only local-AI apps reviewed on this site, Liquid Apollo works on both major mobile operating systems plus Mac and Apple Vision.
- Users comfortable managing an OpenRouter API key. If you already have an OpenRouter account, Liquid Apollo's cloud mode gives immediate access to models like GPT-4 and Llama 3 from a mobile client.
Who Should Not Use Liquid Apollo
- Readers who want a strictly local-only app with zero cloud option. Liquid Apollo's cloud mode is a built-in, first-class feature that cannot be removed from the app, even if you never enable it — Private LLM or Enclave AI are better fits if the app itself must have no cloud code path at all.
- Readers who assume "runs on-device" means every mode is offline. Only the local LFM2 mode is offline; the OpenRouter and custom-backend modes both require a network connection by design.
- Readers who want frontier-scale model capability without setting up OpenRouter or a self-hosted server. LFM2's largest published size is 2.6B parameters, well below the frontier cloud models Liquid Apollo's OpenRouter mode can reach — local mode alone will not match GPT-4-class output quality.
- Windows or Linux desktop users wanting a native client. Liquid Apollo has no Windows or Linux app; only the LM Studio/Ollama server side of custom-backend mode can run there.
- Teams needing centrally managed, audited deployment. Liquid Apollo is a single-user consumer client with no admin console or team billing across its three modes.
Common Mistakes
Frequently Asked Questions
Is Liquid Apollo actually local-only?
No. Liquid Apollo's local mode, using Liquid AI's LEAP technology and LFM2 models, runs fully on-device and offline. But the same app also has an OpenRouter cloud mode and a custom-backend mode, both of which require an internet or network connection. Only local mode is offline.
What is LEAP?
LEAP (Liquid Edge-AI Platform) is Liquid AI's own developer SDK for running small language models directly on a phone or other edge device. Liquid Apollo's local chat mode is built on LEAP, and Liquid AI also offers LEAP separately for other app developers.
What models does Liquid Apollo run on-device?
Liquid AI's own LFM2 model family: LFM2-350M, LFM2-700M, LFM2-1.2B, and LFM2-2.6B parameters, per Liquid AI's published model documentation.
Does Liquid Apollo cost money?
The app itself is free on both the App Store and Google Play, with no in-app purchases listed. Its OpenRouter cloud mode is billed separately by OpenRouter and the model provider you choose, using your own API key — not by Liquid AI.
What is Liquid Apollo's OpenRouter mode?
It lets you paste your own OpenRouter API key into the app to reach hosted models over the internet, including open-source models like Meta Llama 3 and closed-source models like OpenAI's GPT-4. This mode requires an internet connection and is unrelated to Liquid AI's own on-device LFM2 models.
Can I connect Liquid Apollo to my own LM Studio or Ollama server?
Yes. Its custom-backend mode connects to a self-hosted LM Studio or Ollama instance on your own network, letting you use Liquid Apollo as a mobile front-end for a model you run yourself.
What platforms does Liquid Apollo support?
iPhone (iOS 18+), iPad (iPadOS 18+), Mac (Apple Silicon, macOS 15+), Apple Vision (visionOS 2+), and Android via Google Play. There is no Windows or Linux desktop client.
Who makes Liquid Apollo?
Liquid AI, Inc., which also develops the LEAP SDK and LFM2 model family used in the app's on-device mode.
Does Liquid Apollo collect my data?
Liquid Apollo's App Store privacy label states the developer collects no data from the app itself, and local mode needs no account. That claim covers the app, not OpenRouter or a self-hosted backend — those services have their own separate privacy practices once you send data through them.
Should I use Liquid Apollo if I want a fully offline app with no cloud option?
Not ideally. Liquid Apollo always ships with a cloud mode built in, even if you never turn it on. Private LLM and Enclave AI are on-device-only apps with no cloud code path at all, and are better matches for that specific requirement.
Verdict
Liquid Apollo is an honestly-marketed hybrid app: Liquid AI advertises the OpenRouter and custom-backend modes as headline features, not fine print, which makes it easier than most apps in this category to know what you are actually getting. Its local mode is a genuine, free way to run Liquid AI's own LFM2 models fully offline via LEAP, and its cross-platform reach (iOS, Android, Mac, Apple Vision) is wider than most single-platform local-AI apps covered on this site. The trade-off is exactly what the name "hybrid" implies: readers who need an app that is structurally incapable of reaching the cloud, under any setting, should choose Private LLM or Enclave AI instead. Readers who want one flexible client spanning on-device LFM2 chat, OpenRouter-hosted frontier models, and their own self-hosted server get exactly that from Liquid Apollo, at no cost beyond whatever OpenRouter usage they choose to pay for.
Sources
- Liquid Apollo on the App Store — version, rating, price, platform requirements, privacy label, and full app description.
- Liquid Apollo on Google Play — Android availability.
- Liquid AI — company site, LEAP and LFM2 product pages.
- LEAP — Liquid Edge-AI Platform — SDK documentation and model library details.
- Liquid AI Launches LEAP and Apollo — Liquid AI's own launch announcement.
- Liquid AI privacy policy — data-handling statement referenced in the app's App Store privacy label.
- OpenRouter — the third-party service Liquid Apollo's cloud mode connects to.