Your private AI companion. Chat locally with compatible 35B models, connect your tools by choice, and bring ideas to life with images and Seedance video.
A world of your own.
Meet Pumli, your private AI companion for everyday questions, thoughtful research and a little imagination. Start with a model on your phone. Add useful connections when you choose. Make the assistant feel like yours.
BIG AI, RIGHT HERE
Download a compatible model and chat on your device, even offline. Explore MiniCPM-V 4.7 35B-A3B on compatible hardware: 35 billion total parameters, with about 3 billion active per token. Prefer something smaller? Choose a compact model that fits your phone, storage and task.
A WORLD OF LOCAL MODELS
Discover compatible GGUF models through Hugging Face search, alongside curated choices with Hugging Face and ModelScope download sources. Manage downloads, see the models already on your device and remove those you no longer need. Model compatibility and performance vary; not every model file can run on every device.
YOUR PRIVATE LITTLE SPACE
Local inference runs on your device. Keep conversation history close, start a fresh chat, and review or delete what you save. Give your assistant a name, adjust its instructions, and view, edit or remove its locally saved memories. Local conversations do not require a general Pumli cloud chat account.
CONNECT BY CHOICE
Bring notes, code, documentation and search into the conversation with supported connections such as Notion, GitHub, Context7, Microsoft Learn and remote MCP servers. Search the web through the services you enable. Connectors are optional: choose access, review requests and control your tool-approval settings.
A HELPING HAND ON IPHONE
Use supported phone actions for calls, Messages, Contacts, Calendar, Reminders, Maps, Health and Fitness. These features require the relevant permissions and your choice to enable them. Health and Fitness tools are read-only and designed to show results on your device. Phone and messaging actions use the appropriate system interface.
FOLLOW YOUR CURIOSITY
Research mode explores bigger questions across multiple steps using enabled tools. Choose a depth, follow the progress and open the sources behind the answer. Use it to compare ideas, investigate a topic or bring scattered information together. AI research still needs your judgement: check important facts and source material.
FIND THE WORDS
Ask for an email draft, then review and edit the recipient, subject and message. Open the draft in Mail, choose your sending account and send when you are ready. Pumli 3.0 uses this Mail handoff; direct Gmail and Outlook account connectors are not part of this release.
MAKE SOMETHING NEW
Create images with optional online providers, including Gemini, or turn a prompt and supported reference media into a Seedance video. Preview and save supported creations to your device. Use your own eligible provider account or API key; availability, safety rules, quotas and provider fees apply.
SPEAK YOUR WAY
Use local dictation after downloading a compatible speech model, or choose optional online realtime voice. Move between writing, speaking, researching and creating in one personal space.
KNOW WHAT STAYS LOCAL
Online AI, connected search, remote tools, realtime voice and cloud generation require internet access and send the information needed for the task to your chosen provider. Some services require separate accounts, keys or usage fees. Pumli does not promise that online providers retain no data.
Large models need substantial storage and compatible hardware. Speed, context and available features vary by device and region. AI can make mistakes; review important answers and external actions before relying on them.
Pumli is an independent app. Model and service names belong to their respective owners.
This app is incredibly powerful, and the richness of its features is truly astonishing. Thank you for creating such an excellent app! I’ve tried numerous AI chatbots, but this one truly stands out! It seamlessly integrates the best features of Private LLM, ChatGPT, DeepSeek, Claude, Grok, and Gemini. The responses are not only lightning-fast but also incredibly insightful. Whether I’m asking for coding help or just need a chat, this app combines the depth of ChatGPT with the innovative edge of DeepSeek. DeepSeek’s R1 capabilities are deeply integrated, making it a goldmine for knowledge seekers. DeepSeek’s versatility shines through in its ability to handle complex queries with ease. And the privacy aspect, thanks to OneLLM, gives me peace of mind knowing my data is secure. This app is a game-changer for anyone looking for a comprehensive AI experience. Five stars all the way!
Developer Response
Thank you very much for your feedback.
Virtual Box of chocolates
pie guy 0408
I really love the concept of this app. Unfortunately it’s so inconsistent. I have gone a little more than a week with multiple uses of this app. I’ve added most API and used roughly 50% of the features. Everything was fantastic and then there’s an update. I think “Oh, great new feature, more securities, bug fixes, stability improvements etc.” I use the Online LLM’s and receive a exceed daily use. I didn’t use the app for a couple of days. Okay so I close, restart device. Open app and receive API error. Fine, I delete the current number get all new ones. I get warning that I need to add API. I close and restart device. Still not working. I try local LLM and app works flawlessly. I tried online LLM a few hours later and it worked. Two days later nothing worked. Out of curiosity I tried the non pro model version and it worked great. The following day Pro worked great. Today the Pro only generates images via Gemini and that’s it. Again I really want to love OneLLM Pro I just prefer opening OneLLM Pro minus the virtual Box of chocolates. I don’t even know anyone named Forest. I do like shrimp though. My favorite is popcorn shrimp, shrimp Alfredo, shrimp linguine, cocktail shrimp, grilled shrimp piccata skewers, garlic shrimp anything, shrimp po’ boy…
Developer Response
Thank you for your feedback. The Gemini API is not entirely free, as Google provides a certain level of usage for free-tier users. However, to access features such as Gemini video generation, image generation, and the 2.5 Pro LLM, you would need to upgrade from a free-tier user to a paid user on the Google platform. Please note that free-tier users may experience throttling and usage frequency restrictions, which are unrelated to the app’s stability. Any error messages displayed by the app will only reflect the messages received from the Gemini API.We sincerely appreciate your valuable feedback. If you encounter any issues while using the app, please feel free to contact us on X or BlueSky, or you can also send an email to: hello@1stdo.com
Perfect App
Benjamin169
This app is fantastic! It packs a ton of features and comes at a super affordable price. It brings together almost every model out there—ChatGPT, DeepSeek, Mistral, Grok, Claude, you name it—and supports both AI images and AI videos. It's incredibly versatile, and I can even use it to connect to Ollama. However, there's definitely room for improvement. I'd love to see some user guides or tutorials from the developers because there are so many features that a manual would be really handy.
Developer Response
We sincerely appreciate your valuable feedback. If you encounter any issues while using the app, please feel free to contact us on X or BlueSky, or you can also send an email to: hello@1stdo.com
Pretty great.
syscheckin
Definitely more feature-focused than other local AI apps. It would be 5 star for sure if they added local transcription by offering options to download the various transcriptions models available like Whisper which work with iOS. It already sort of discerns itself well as offering a lot more than other apps, though a greater focus on local/non-cloud features in the present and future (as devices get more capable) would really solidify the value as it does have a higher price tag. It would be really fun to somehow load 2-3 <1b local models and have them chat with each other like the existing feature.Also, according to the app store listing; thank you for trying to not do what so many other apps do with data collection and layered third party involvement.
Developer Response
Thanks for your feedback and comments! We’ll definitely consider your suggestions for adding more transcription models. Moving forward, we plan to provide more model download options in the app. As for the multiple-model chat feature you mentioned, that sounds super interesting, and we’ll definitely look into it further! Thanks again for your input. In our app, local LLMs are processed by local engines, and online LLMs are hosted by users themselves—we don’t collect any data!
Meet Pumli 3.0 — a more personal space for local AI.
• A redesigned assistant with conversation history and editable preferences.
• On-device model choices, including MiniCPM-V 4.7 35B and compatible models.
• Multi-step Research with progress, sources and adjustable depth.
• Optional connectors with clearer setup and tool approval controls.
• Editable email drafts before opening Mail.
• Online image creation and Seedance video tools using your own provider account or key.
• Refined layouts for larger displays and improvements to everyday interactions.
Version 3.0.1
The developer, ESOTECH Ltd, indicated that the app’s privacy practices may include handling of data as described below. For more information, see the developer’s privacy policy .
Data Not Collected
The developer does not collect any data from this app.
Privacy practices may vary, for example, based on the features you use or your age. Learn More
The developer indicated that this app supports the following accessibility features. Learn More
Supported Features
Voice Control
Dark Interface
Differentiate Without Color Alone
Sufficient Contrast
Reduced Motion
Information
Seller
ESOTECH LTD
Size
181.7 MB
Category
Productivity
Compatibility
Requires iOS 16.4 or later.
iPhone Requires iOS 16.4 or later.
iPad Requires iPadOS 16.4 or later.
Mac Requires macOS 12.0 or later and a Mac with Apple M1 chip or later.
