Treena launches an iPhone coding environment that runs agents without a laptop

Why it matters

Treena is testing whether AI agents can turn the phone into a primary development interface. Success depends on developer trust in its cloud workspace and broad GitHub permissions.

Camden Hulse (@camden_hulse) and Kollaikal Rupesh (@KollaikalRupesh) launched Treena on July 30th, an iPhone coding environment designed to start AI agents on a GitHub repository, run the development workspace in the cloud and push finished changes without a laptop.

https://x.com/camden_hulse/status/2082874643095195800?s=46

Hulse announced the release in a thread on X, describing Treena as a coding environment with "no laptop in the loop." The iOS app is live in Apple's US App Store. Hulse said the Android version was still awaiting Google Play approval at the time of the launch.

The product grew out of a constraint Hulse encountered away from his desk. Hulse, who says he is 19, wrote in May that he dropped out of the University of Florida after his mother died. A week later, he needed to ship software from a doctor's office and had only his phone. He named Treena after his mother and moved to San Francisco to build it at The Residency, a live-in program that provides builders with housing, workspace and a full-time cohort. Treena describes itself as backed by The Residency.

Rupesh brings a different engineering background to the project. Before Treena, he worked on real-time voice AI at Wayline and Smallest.ai. He earned a master's in business analytics from the UC Davis Graduate School of Management, where he served as president of the Directors' Student Advisory Committee, after studying information technology at Saveetha School of Engineering.

The phone becomes the control surface

Treena asks a user to sign in with GitHub, select a repository and provision a dedicated cloud workspace. From the phone, the developer can give an agent a task, review its plan, inspect files and diffs, use a Linux terminal, run tests, commit the result and push it to GitHub.

The App Store listing says Treena supports Claude Code and Codex, along with direct model access through user-supplied OpenRouter, OpenAI or Anthropic keys. The app also includes a code editor, Git controls, browser previews, voice input, secrets management and Model Context Protocol connections. Agents continue running after the app closes, with push notifications when work finishes or requires approval, according to Treena.

That architecture shifts most computation away from the device. Treena provisions the workspace remotely, while the phone handles instructions, review and approval. The founders described the approach in a July 20th blog post as a full mobile IDE rather than a companion screen for a desktop session. A repository can be opened cold from the phone, without an existing laptop process to resume.

Treena's initial release uses a bring-your-own-account or bring-your-own-key model for AI access. The App Store lists a free tier and paid plans for larger workspaces, concurrent agents and longer unattended runs. Pro costs $19.99 per month or $191.99 per year. Max costs $99.99 per month or $949.99 per year.

Repository access carries a security tradeoff

Running an agent against production code from a phone requires broad access to repositories and credentials. Treena's privacy policy says its GitHub authorization receives read and write access to all public and private repositories because GitHub does not offer a narrower OAuth scope for this workflow. Treena says it acts only on repositories a user opens.

Repositories are cloned onto a dedicated Fly.io virtual machine. Treena stores connected Claude and Codex credentials on its servers and inside the workspace so those engines can run under the user's account. Direct API keys are stored in the device keychain and sent straight to the selected model provider, according to the policy. Agent transcripts and cloned repositories remain on the workspace volume, while Treena says transcripts are pruned after 45 days.

Those permissions make Treena useful for real development work and raise the threshold for trust. A mobile coding interface can be lightweight; the cloud machine behind it still handles private source code, credentials and write access to GitHub. Treena will have to make that operational model legible to developers accustomed to keeping sensitive repositories inside established desktop and cloud development systems.

A contested category

Treena is entering a market where AI agents have already weakened the connection between writing software and physically operating a desktop IDE. CloudCLI also runs coding agents in persistent cloud environments that can be started from a phone, browser or desktop editor. Products such as Lunel and Shunt are pursuing adjacent mobile and remote agent workflows, while Replit and GitHub Codespaces already provide broader cloud development environments.

Treena's bet is narrower: mobile should serve as the primary place to start and finish a coding task, including the editor, terminal, tests, diff and Git push. The founders are building around moments when a developer needs to fix an incident, test an idea or supervise an agent before reaching a laptop.

AI agents make that interface plausible because the user no longer has to type every line on a small touchscreen. The harder test begins after launch: whether developers want to conduct enough of the review, debugging and approval loop on a phone to make a separate mobile IDE part of their regular toolchain.