What it does
Puma Browser can run local language models on a phone and also offers access to cloud models, alongside its decentralized-web browsing features.
Puma offers a choice between local and cloud AI on iOS and Android. The provider says supported Gemma, Qwen and Ministral models can run on the phone, including offline; its policy says local model chats are not sent to Puma or cloud AI providers. Optional OpenAI, Anthropic and Gemini features send the content submitted for those requests to the chosen provider.
Local chat storage is only one part of the data picture. Puma's policy also lists app usage and device information associated with randomly generated IDs, plus Sentry crash reports when allowed. Retention depends on purpose and legal requirements rather than one fixed period. The mobile app is free; a Mac TestFlight route is described as forthcoming.
Strengths
- On-device model option
- Offline local-chat workflow
Best for
- Mobile users comparing local and cloud AI in a browser
Sources
- Puma Browser — mobile browser for AI agents. Private by design. (Oct 26, 2026)
- Puma Browser / Privacy Policy (Oct 26, 2026)