Frequently Asked Questions

TokForge is a private AI chat app for iPhone and Android that runs large language models (LLMs) entirely on your device. It supports character cards, thinking/reasoning mode, dual inference backends, and built-in benchmarking, all without requiring an internet connection.

No. Once you've downloaded a model, TokForge works 100% offline by default. Internet is only needed to download models from Hugging Face, and for opt-in web search if you choose to turn it on (it's off by default). All AI inference runs locally on your phone's CPU or GPU.

Never. TokForge has no analytics, no telemetry and no cloud inference, and we receive none of your data. Your conversations, characters, and settings stay on your device. The optional benchmark sharing feature is entirely opt-in and only shares device specs + performance numbers, never conversation data.

Yes, TokForge is completely free. There are no in-app purchases, subscriptions, or ads. Future pricing has not been determined.

Yes. There's a full native iPhone & iPad build (iOS 16+) at near feature-parity with Android: on-device chat, group chat & roleplay, characters & personas, image generation (Apple CoreML / Neural Engine), vision, voice, RAG, and the agent suite. It's in public beta on TestFlight: join the iOS beta. Everything runs on-device, same as Android.

The Android app interface ships in 8 languages: English, French, German, Spanish, Russian, Chinese, Korean, and Turkish. It follows your system language, or you can pick one in Settings. The iPhone and iPad beta is currently English only. The language the AI can chat in is separate: it depends on the model you download.

TokForge runs on iPhone & iPad (iOS 16+) and any ARM64 Android device running Android 8.0+. For a good experience with larger models, you'll want a recent A-series / Snapdragon-class chip and at least 8GB RAM.

We test on phones and tablets from Samsung, Google Pixel, Xiaomi, OnePlus, RedMagic and Lenovo, plus iPhone and iPad, with more devices added over time.

It depends on the model size:

4GB RAM: the smallest models (about 1B and under) run and stay responsive

8GB RAM: around 4B is the everyday sweet spot

12GB RAM: 8B to 9B models, with headroom for longer chats

16GB and up: 12B to 14B models join the list

Speeds move with the app version, the model and the handset, so we do not freeze numbers into this page. The phone-speed leaderboard carries live results from real handsets, per device and per model.

TokForge auto-detects your RAM and recommends the right model size during setup.

Model sizes vary by quantization:

Smallest models (about 1B and under): roughly 0.6 to 0.8GB

Around 4B: roughly 2 to 3GB

Around 8B to 9B: roughly 5 to 7GB

Around 14B: roughly 10GB

Exact size depends on the format (GGUF or MNN) and the quantization, and the app shows it before you download. You can download and delete models freely. The app download itself is a little over 100MB.

Yes! The iOS build is universal (iPhone & iPad), and on Android TokForge has a responsive layout that adapts to tablet screens.

TokForge has two inference backends:

MNN: Runs on the CPU, with OpenCL and Vulkan GPU paths depending on the phone. Most curated chat models ship in this format.

GGUF (llama.cpp): Runs on the CPU, with OpenCL and Vulkan GPU paths. This is the format most models on Hugging Face come in, so it is the one to use for your own imports.

TokForge picks a default for your device and model, and AutoForge can test the options on your own handset. Both formats support thinking mode on reasoning models.

Yes. TokForge includes a built-in Hugging Face model search, so you can find and download any compatible GGUF model directly from the app. There are also 52 curated models in the Android catalogue (and a separate iPhone catalogue), filtered to what your phone can run.

Some models (like Qwen3, DeepSeek-R1, and Phi-4 Mini Reasoning) can "think out loud" before answering, showing their chain-of-thought reasoning inside <think> blocks. TokForge renders these as collapsible sections so you can see how the model arrived at its answer. It works with both MNN and GGUF models that support it.

TokForge supports the TavernAI V2 character card specification. You can import cards as PNG files (with embedded metadata) or JSON. Features include lorebooks, alternate greetings, world info, {{char}}/{{user}} placeholders, and system prompt assembly.

Yes. TokForge ships with four built-in personas: Rex (direct and analytical), Luna (creative and warm), Marcus (scholarly and thorough), and Aria (artistic and expressive). You can use these as-is or as templates. The persona system also lets you build a reusable persona library, bind a persona per conversation, and import personas and whole chat histories from SillyTavern, Layla, and PocketPal.

Yes. TokForge generates images on-device with Stable Diffusion: SD1.5 and LCM models, SDXL models on phones with enough memory, and an optional NPU tier on supported Snapdragon phones. You can lock a face or subject across images with reference-image identity (IP-Adapter). No cloud, no API fees, and it works in airplane mode.

Yes. Attach a photo and ask about it. On Android, Qwen3.5 and Gemma 4 E2B models read images natively, and a small SmolVLM helper describes photos for models without built-in vision. Gemma 4 E2B on Android also takes audio and video input. On iPhone, the app describes the photo on-device and passes that description to the chat model. All of it runs on the device.

Yes. Build a reusable persona library and bind one per conversation, then use the dedicated roleplay mode with lorebooks, multi-character group chat, and rotation. You can import characters and whole chat histories from SillyTavern, Layla, and PocketPal.

Optionally. Web search in chat is opt-in and off by default, and it's consent-gated each time. TokForge stays fully offline unless you explicitly turn it on.

Agents let you build no-code, multi-step workflows that call tools and run entirely on your device, with bounded loops and triggers. TokForge ships with 12 built-in agents and a simple builder, and your own agents can call an on-device image-generation tool.

Tokens per second (tok/s) measures how fast the model generates text. A token is roughly 3/4 of a word. For comfortable reading speed, you want at least 5-8 tok/s. Above 15 tok/s feels nearly instant. TokForge shows tok/s in real-time during conversations and in AutoForge results.

AutoForge is TokForge's built-in benchmarking and tuning tool. It measures how fast models run on your own phone (decode speed, prompt processing and time to first token), tries the settings that apply to your model, and suggests the fastest setup, which you apply with one tap. Results stay on the phone, and you can export them or choose to submit them to the public leaderboard.

On Android, TokForge is live on the Google Play Store. On iPhone & iPad it's in public beta on TestFlight: join here. You can also grab either from the homepage.

Yes. TokForge is live on the Google Play Store for Android (version 1.3.6.1), so anyone can install it. The iPhone and iPad build is a free public beta on TestFlight and is not on the App Store yet.