@Private Provider App successfully writing code with Pi Agent in a TEE with open weight models via @routstr, packaging an apk, signing android key and publishing an APK to blossom via #ngit all from inside the app on a GrapheneOS pixel 6a.

Who runs a #blossom server that allows APK's? #asknostr

@Private Provider App coming soon ™️ View quoted note →

I honestly am shocked by what I built. I wasn't even sure it was possible, but here I am with @routstr and Pi Agent on my GrapheneOS pixel 6a phone in a minimalist ux. I have all the top models and a harness, writing code and shipping all from within the app without any permission. I built the tool that I felt was missing from being GrapheneOS only. I can now vibe code from anywhere, at anytime, completely private without a laptop. @Private Provider App releasing soon ✅

@routstr is in out in the wild, you can't stop us 🌐⚡🪄 View quoted note →

One of my favorite features I built into @Private Provider app is the visualization per prompt of the actual TEE attestation steps @routstr is performing once a user hits send. It reads the attestation calls, encryption steps and shows them as they actually happen. Truth drives the animation. Then I provide complete transparency on what data the app used and offer users a way to verify independently.

I built a multi panel android AI app. Five distinct tabs make up the app: chat, pi agent, workspace, wallet and monitoring dsshboard. The app took @routstr daemon, pi agent, #ngit + #amber and runs them in the android developer environment and emulates them via proot. This gives Android users full functionlality of these tools in a single clean app. Users can select Tinfoil open weight AI models in the app, have their prompts encrypted on device before being processed in an attested TEE and returned. This is truly private mobile AI. Encrypted routing and blind to the AI providers, not just in travel like other "private ai apps." When using Tinfoil models users prompts are never seen. Users can also downgrade to frontier models that routstr still proxy for you, the provider doesn't know who you are but prompts are still in clear text to the AI upstream. Secure prompts are shown with a green attestation check, while unsecure prompts highlight a yellow warning and require user approval to downgrade privacy. The chat window can have a unique model dsitinct from the pi agent window. Users can save, switch and start new chats or pi sessions at any time. Pi Agent on the phone reads, writes, edits etc and all outputs go to the app's dedicated workspace panel. I've include a simple ngit cli tool that guides new users through a new project development. Once a project is made users can select directories from workspace and create ngit repos, publish to grasp servers, commit, push and copy ngit urls. Users can clone ngit repos and work on them. Amber ngit support is built in but not required. An in app wallet is presented to make topups as easy as paying a built in npc wallet address, bolt11 invoice or a cashu token. The wallet has full functionality. The monitor window represents a complete dashboard of every token spend, every model used, providers, clients, average spend etc to fine tune and document your actions. All this runs on a GrapheneOS device. No account, no subscription, no email, no phone number, no photo id. Pay the actual compute costs per query privately in #cashu, choose from frontier or open weight, choose secure TEE enclave or not, switch between sessions and chats, build with Pi Agent on the phone, push with ngit and sign with Amber, receive and send with wallet and monitor the entire situation. This app lets you be on the move, access AI privately at any time, build anything anonymously, publish to ngit anywhere, all while doing so from a secure GrapheneOS device! First release coming soon ™️

My @Private Provider App has the @routstr monitor in it!

@Private Provider an Android @routstr App🔥

I made a @Private Provider Android @routstr app! Chat + Pi Agent + Workspace +Wallet + Monitoring All on an android GrapheneOS! Private by default.