fullmoon
Fullmoon runs small Llama and DeepSeek models entirely on your Apple device — free, offline, open source.
Fullmoon is an easy recommendation if you are on Apple hardware and your priority is that nothing you type leaves the device. It is free, open source, and the install is genuinely simple: pick one of four bundled models (Llama-3.2-1B/3B at 4-bit, DeepSeek-R1-Distill-Qwen-1.5B at 4-bit or 8-bit), let it download, and you are chatting offline. The ceiling is real, though — 3B parameters at 4-bit will not carry multi-step reasoning or long documents, and there is no Windows or Android build. If you need that kind of horsepower you are looking at a cloud model or a heavier local stack such as Ollama or LM Studio on a desktop with more memory. Treat Fullmoon as the private, pocket-sized end of
Verified 4d ago · liveness 58/100 · cite: rightaichoice.com/tools/fullmoon
- Privacy-focused Apple users
- Offline and air-gapped use cases
- Developers evaluating small quantized models
- Students and researchers needing a private offline assistant
- Windows or Android users
- Workloads needing frontier-level reasoning or very long context
- Teams requiring shared workspaces and multi-user collaboration
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Fullmoon if you need frontier-level reasoning, long-context document work, or a Windows/Android build — the bundled models top out at 3B parameters and the app is Apple-only.
Larger models cost disk space on your device: the 3B 4-bit build is 1.8 GB and the 8-bit DeepSeek build is 1.9 GB, so a full set of four eats close to 5.4 GB of storage.
Fullmoon is free and open source, which puts it at the bottom of the cost curve against paid cloud assistants and paid local-LLM GUIs. The tradeoff is capability, not price: you are trading frontier-model reasoning and long context for zero inference bills and nothing leaving your device.
In short
fullmoon — Fullmoon runs small Llama and DeepSeek models entirely on your Apple device — free, offline, open source. Best for Privacy-focused Apple users, Offline and air-gapped use cases, Developers evaluating small quantized models. Free to use.
What people actually say about fullmoon — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
59 mentions across 5 sources (Hacker News, YouTube, Product Hunt, GitHub, Lemmy) · researched Jul 3, 2026.
Average across the 5 sources that answered — each source counts once, not each post.
- +Fully offline and private—no data leaves your device.
- +Optimized for Apple Silicon with Metal 3 and MLX.
- +Clean, simple interface that even beginners can use.
- +Free and open-source with no hidden costs.
- +Supports 4-bit and 8-bit quantization to reduce memory.
- −Naming confusion with another open-source 'fullmoon' project.
- −Limited model selection—only three small models bundled.
- −No API access for remote or automated use.
- −No RAG capability—can't query local documents yet.
- −Small community means slower support and fewer contributions.
- • No hidden costs—the app is completely free.
Viability Score
How well maintained and how widely used is fullmoon? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Runs LLMs fully on-device with no network connection
- Swift MLX inference framework
- Apple silicon optimization with Metal 3 graphics
- Four pre-quantized bundled models (4-bit and 8-bit)
- Llama-3.2-1B-Instruct-4bit (0.7 GB)
- Llama-3.2-3B-Instruct-4bit (1.8 GB)
- DeepSeek-R1-Distill-Qwen-1.5B-4bit (1.0 GB)
- DeepSeek-R1-Distill-Qwen-1.5B-8bit (1.9 GB)
- Customizable chat theme and fonts
- Custom system prompt
- Apple Shortcuts action for local-model output
- Runs on iOS, iPadOS, macOS, and visionOS
- Open-source codebase (fullmoon-ios.git)
- Free and open source
- TestFlight builds for latest features and models
About fullmoon
Fullmoon is a free, open-source app from Mainframe that runs large language models entirely on your Apple hardware. After an initial model download it works with no internet connection at all, so your prompts and responses never leave your iPhone, iPad, Mac, or Vision Pro. It is built on Swift MLX and tuned for Apple silicon with Metal 3, and it ships four quantized models you can switch between: Llama-3.2-1B-Instruct-4bit (0.7 GB), Llama-3.2-3B-Instruct-4bit (1.8 GB), DeepSeek-R1-Distill-Qwen-1.5B-4bit (1.0 GB), and DeepSeek-R1-Distill-Qwen-1.5B-8bit (1.9 GB). You can restyle the chat with your own theme and fonts, write a custom system prompt, and pipe local-model output into the rest of your phone or Mac through Apple Shortcuts. It is aimed at privacy-conscious Apple users, people who work without reliable connectivity, and anyone who wants to poke at small quantized models without paying cloud inference bills. What it is not: a cloud assistant. The model list tops out at 3B parameters and the app is Apple-only, so heavy reasoning, long context work, and Windows or Android users belong elsewhere.
Behind the Verdict
Fullmoon's pitch is narrow and it executes on it. The whole app is a local inference shell for Apple platforms: Swift MLX under the hood, Metal 3 acceleration, and four pre-quantized models that Mainframe has already converted so you never touch a conversion script. The tech specs page is unusually candid about what you are downloading — parameter counts, tensor types (FP16 • U32), precision, base model, and file size for each entry — which is the kind of detail that tells you the team actually ships inference rather than a wrapper around someone else's API. Where it shines: offline. Once a model is on disk the app runs with no connection, which matters on flights, in the field, in secure facilities, or anywhere you would rather not send a prompt to a server. Personalization is deeper than most free chat apps — theme, fonts, and a full custom system prompt — and the Apple Shortcuts action means you can wire a local model into automations alongside your other actions rather than living inside one chat window. The honest limits are the models and the platform. The largest option here is Llama-3.2-3B-Instruct at 4-bit, 1.8 GB on disk. That is fine for short drafting, summarizing a paragraph, quick lookups, and playing with a reasoning-distilled model like DeepSeek-R1-Distill-Qwen-1.5B. It is not fine for long-context analysis, codebase-scale work, or anything where you would normally reach for a frontier cloud model. There is also no path off Apple hardware: iOS, iPadOS, macOS, and visionOS only, so a mixed Windows/Android team cannot standardize on it. And because everything is on-device, quality scales with the device — an older iPhone will feel the 3B model more than an M-series Mac will. Where it fits: individual Apple users who want a private assistant that never phones home, students and researchers who need an offline fallback, developers who want a fast way to compare small quantized models on real Apple silicon, and anyone demoing on-device AI without a cloud bill. Where it does not: teams that need shared workspaces and collaboration, anyone whose workload depends on large context or frontier-level reasoning, and non-Apple users entirely.
Researching fullmoon? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas fullmoon actually fits — and what changes day-one when you adopt it.
Downloads Llama-3.2-3B-Instruct-4bit (1.8 GB) on Wi-Fi at home, then uses Fullmoon in airplane mode to summarize notes and draft replies during the flight.
Outcome: Answers and drafts without a connection and without any prompt leaving the phone.
Loads the 4-bit and 8-bit DeepSeek-R1-Distill-Qwen-1.5B builds side by side and runs the same reasoning prompts to compare output quality against the extra 0.9 GB of disk the 8-bit version costs.
Outcome: A grounded read on the quality/size tradeoff on their own Apple silicon before committing a model to a project.
Sets a custom system prompt for tone and wires the Apple Shortcuts action into a workflow that takes selected text and returns a rewritten version from the local model.
Outcome: Rewriting lives inside existing Mac shortcuts, with the text never going to a server.
Use Cases
- Chat with a private AI assistant on a flight or anywhere without a connection.
- Draft and reword short text on your iPhone or Mac without sending it to a cloud server.
- Compare latency and output quality of 1B-3B quantized models on your own Apple silicon.
- Wire local-model output into Apple Shortcuts automations alongside your other actions.
- Demo on-device AI to a class or client with no internet and no inference bill.
- Keep sensitive notes or prompts on-device when policy forbids sending them off-device.
Models Under the Hood
as of 2026-10-16
Limitations
- Fullmoon only runs on Apple platforms — iOS, iPadOS, macOS, and visionOS — so Windows and Android users are out.
- Inference happens on-device against a fixed set of four bundled quantized models, the largest being Llama-3.2-3B-Instruct at 4-bit (1.8 GB), which rules out long-context analysis and heavy multi-step reasoning.
- You need an initial download per model before you can go offline: 0.7 GB for Llama-3.2-1B-4bit, 1.8 GB for Llama-3.2-3B-4bit, 1.0 GB for DeepSeek-R1-Distill-Qwen-1.5B-4bit, and 1.9 GB for the 8-bit variant.
- Output speed and quality track your device's Apple silicon and available memory rather than a datacenter.
- There is no built-in collaboration or shared workspace, and the app is single-user by design.
as of 2026-10-03
Verification history
We have re-verified fullmoon 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where fullmoon's pricing actually pencils out — and where peers do it cheaper.
Fullmoon is free and open source, which puts it at the bottom of the cost curve against paid cloud assistants and paid local-LLM GUIs. The tradeoff is capability, not price: you are trading frontier-model reasoning and long context for zero inference bills and nothing leaving your device.
Setup time & first value
How long it actually takes to get something useful out of fullmoon — broken out by persona, not the marketing-page minute.
Install from the App Store and you are chatting in a couple of minutes, but budget download time per model: 0.7 GB for Llama-3.2-1B-4bit, 1.8 GB for Llama-3.2-3B-4bit, 1.0 GB for DeepSeek-R1-Distill-Qwen-1.5B-4bit, 1.9 GB for the 8-bit build. TestFlight is the faster route to new models.
Switching to or from fullmoon
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a cloud chat assistant: move the prompts you use most into Fullmoon, set a custom system prompt to match the tone, and keep the cloud tool for tasks that need a larger model.
- →From a desktop-only local runner: install Fullmoon on iPhone or iPad so the same offline workflow travels with you.
- →From Apple Shortcuts text actions: swap the action to Fullmoon's shortcut so output comes from the on-device model.
- ↗To a cloud model: if you outgrow 3B parameters, keep Fullmoon for offline and private tasks and route heavy reasoning elsewhere.
- ↗To a desktop local-LLM stack: move up when you need bigger weights than the 3B ceiling Fullmoon ships.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “fullmoon”, and we withheld 6: 6 could not be judged, because “fullmoon” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about fullmoon.
Official links
Tools that pair well with fullmoon
Common stack mates teams adopt alongside fullmoon, with the specific reason each pairing earns its keep.
Atomic Chat
Atomic Chat runs 1,000+ local LLMs offline on your own hardware — free, open-source, no account, no rate limits.
Kai
Open-source, cross-platform AI assistant that generates full interactive screens and runs locally — with persistent memory and an autonomous heartbeat.
Cortex.cpp
Run local HuggingFace models or route prompts to Claude, GPT and Gemini with your own API keys
Featured Head-to-Head Comparisons
Fullmoon vs Spider Cloud
Fullmoon and Spider Cloud serve entirely different needs. Fullmoon is ideal for Apple users who want private, offline local LLM chat with no cost. Spider Cloud is a paid cloud API for developers who need fast, reliable web data extraction for AI agents and RAG pipelines. Choose based on whether your priority is on-device privacy or web-scale data ingestion.
Fullmoon vs Voyage Ai
Choose Voyage AI if you need high-accuracy embeddings and rerankers for enterprise RAG with domain specialization (finance, legal) and compliance; choose Fullmoon if you want a free, private, local LLM chat on Apple devices. They serve completely different needs.
Fullmoon vs Temporal Ai
Choose Temporal AI if you need rock-solid orchestration for multi-step AI agents or microservices that must survive crashes and scale. Fullmoon is the choice for private, offline LLM chat on Apple devices, ideal for privacy-conscious users who only need small models (≤3B). They serve completely different needs – Temporal is enterprise infrastructure, fullmoon is a local chat app.
Alternatives to fullmoon
View all
Atomic Chat
Atomic Chat runs 1,000+ local LLMs offline on your own hardware — free, open-source, no account, no rate limits.
Kai
Open-source, cross-platform AI assistant that generates full interactive screens and runs locally — with persistent memory and an autonomous heartbeat.
Cortex.cpp
Run local HuggingFace models or route prompts to Claude, GPT and Gemini with your own API keys
Frequently Asked Questions
Used fullmoon? Help shape our editorial sentiment research.