Agentic coding on mobile: plan, act, ship

Chatbots can suggest code. Agents can change systems.

That distinction is why mobile finally works for real software work. A chat transcript is not an IDE. An agent with tools inside a workspace is.

Plan mode vs Agent mode

Treena exposes two deliberate modes:

Plan mode

The agent drafts a step-by-step approach. You approve before anything is written. Ideal when:

• You are new to the repo
• The change touches auth, billing, or data
• You want a mental model before execution

Agent mode

The agent edits, runs commands, and iterates until the task is done (or blocked). Ideal when:

• The task is well-scoped
• You trust the surrounding tests
• Speed matters more than ceremony

One toggle. Same conversation. Different risk profiles.

Tools beat pure text

A useful mobile agent must be able to:

• Read and search files
• Apply patches
• Run the terminal
• Observe errors and try again
• Produce a reviewable diff

Without tools, you are copy-pasting into a chat window. With tools, the phone becomes a remote operations console for your codebase.

Bring your own key

First release philosophy: your model, your credentials.

Connect OpenRouter, OpenAI, Anthropic, or account-linked flows like Codex and Claude Code. You keep control of cost, policy, and provider choice. Treena is the environment and the loop, not a locked model tax.

What "done" means on mobile

Done is not "the agent said something smart." Done is:

1. Tests green (or a clear reason they are not)
2. Diff reviewed
3. Commit created
4. Push to GitHub

If those four can happen on a phone, mobile is not a niche. It is a first-class development surface.

The future is pocket-scale

Software is becoming less about sitting still and more about being available when the idea (or the incident) arrives. Agentic mobile IDEs close the gap between "I noticed it" and "I shipped the fix."

That is the future Treena is building toward.

Join the waitlist, and keep your laptop optional.