Four-panel mobile navigation
Before this update the mobile IDE in easy mode showed one thing at a time — the chat — with no direct way to jump to your files, browse the code editor, or switch to the live preview without leaving the current context. That is fixed.
The project header now shows a four-button toggle at the top of the screen on mobile:
- Chat — the AI conversation panel (where you always land by default)
- Files — the full file tree, so you can browse, rename, or delete files without a workaround
- Code — the code editor, with full syntax highlighting and editing
- View — the live preview of your running app
Switching between panels is instant — no reload, no lost state. The AI keeps streaming in the background while you browse your files or check the preview. The bottom dock mirrors this navigation so you always have one-tap access from anywhere in the IDE.
One important detail: in easy mode the dock tabs no longer toggle the "AI enabled" sidebar state that the full desktop IDE uses. Each tab simply switches the active view. This removes a whole class of accidental state flips that were happening when users tapped between panels on mobile.
Message queue during AI responses
Previously, typing and sending a message while the AI was still responding either did nothing or interrupted the current generation. Neither is the right behaviour when you are on mobile and thinking ahead.
Now the input box stays fully active during any AI stream — whether that is a normal chat response or a multi-step pipeline run. Messages you send during a response are queued and shown in a small badge at the bottom of the chat panel. The moment the current response finishes, the next queued message fires automatically.
The placeholder text changes to "Type to queue next message…" while a response is in progress, so the behaviour is always clear. If you queue multiple messages, the badge shows the count. You can expand the queue to review or cancel individual messages before they send.
This change also covers pipeline runs specifically — a gap that existed before where messages sent during an active pipeline run were silently dropped rather than queued.
Smarter pipeline auto-preview
When a pipeline finishes on mobile, the IDE used to automatically switch you to the View panel. This sounds convenient but caused two problems:
- If you stopped the pipeline mid-run, you still got shunted to the preview — which showed a half-built site.
- If the pipeline ran a read-only task (analysing code, generating a report) with no file changes, switching to the preview was pointless.
The new behaviour: the IDE only switches to the View panel when both conditions are true — the pipeline ran to completion without the user stopping it, and at least one file was written, edited, or deleted during the run. If you stop a run or the pipeline touches nothing, you stay exactly where you are.
Inline cost & timing chip
Cost and timing information used to live in a separate expandable card near the bottom of the chat, disconnected from the message it referred to. On mobile that card was easy to miss, and on desktop it added visual noise to the input area.
Both are gone. In their place, each AI response now shows a small chip directly below the message bubble. It displays:
- The total response time in seconds
- The cost in USD once it has been confirmed by the server
- "Cost loading…" while the final figure is still being fetched
This applies to both regular chat responses and pipeline runs. Pipeline costs in particular went through a significant fix: previously the cost chip would briefly flash $0.00 before updating to the real figure once polling completed. The server now emits a definitive cost event at the very end of a pipeline run, so the chip goes straight from "loading" to the correct number — no flash, no intermediate zero.
If you stopped the pipeline before it finished, the chip still captures whatever cost was incurred up to that point rather than showing nothing.
Cleaner mobile chat screen
Several elements that were visible on mobile but offered no useful interaction on a small screen have been cleaned up:
- The IDE footer is hidden on mobile. The footer contains links that belong on desktop — it took up space and added scroll length on phones without adding value.
- "Spawning agents…" and "Gathering context…" animations are desktop-only. These animated status cards appeared during pipeline runs and during context-gather steps. On a phone screen they pushed content down without giving any actionable information. They are now hidden on mobile and remain visible on desktop where screen real estate allows.
- Several input bar controls are desktop-only. Tool toggles (browser mode, file attach, pipeline toggle, agent mode button) are hidden on small screens where they were cramped. The core send/mic flow remains fully functional on mobile.
Thinking card now shows seconds
When you use a reasoning model (Claude with extended thinking, o3, QwQ, and similar), the AI's hidden reasoning block is shown as a collapsible "Thinking" card above the final response. The card previously showed "Thought for 42w" — a word count of the internal reasoning — which was not meaningful to most users.
It now shows "Thought for 8s" — the actual wall-clock time in seconds from when the message was sent to when the reasoning block closed. For long reasoning runs it switches to minutes: "Thought for 2m 14s". This gives you a real sense of how much compute the model spent on the problem.
Everything at a glance
| Change | Before | After |
|---|---|---|
| Mobile navigation | Chat only in easy mode | Chat, Files, Code, View |
| Typing during AI response | Message dropped or interrupts | Queued, fires after response ends |
| Pipeline auto-preview | Always jumps to View on finish | Only if run completed and files changed |
| Cost display | Separate expandable card | Inline chip under each AI message |
| Pipeline cost flash | Shows $0 then real cost | Goes straight to final cost |
| IDE footer on mobile | Visible | Hidden |
| Pipeline status animations | Shown on all screens | Desktop only |
| Thinking card duration | Word count (e.g. "42w") | Seconds (e.g. "8s") |
All of these changes are live now — no update required on your end. Open the IDE on your phone to see the new navigation and queue badge immediately.
Related: How to automate your dev workflow with the Pipeline Builder · Total control over your AI agent · Claude Fable 5 on AICODESIT