August brought ISEA closer to the point where more people can actually use it every day. The Android beta now has a production package and a clearer download path, platform-specific glass surfaces are taking shape, and the Agent feels quicker to operate in both voice and long conversations.
We also worked on the recovery paths around the app: pulling a complete sync, applying remote configuration, restarting after an update, and reporting feedback from inside a conversation. These are the details that make a mobile product feel available rather than fragile.
Android is ready for a wider beta path
The Android build pipeline produced a production APK, and the landing site now routes downloads by platform so Android users reach the right release asset. The beta download is labeled more clearly, which makes the handoff from the website to the app less ambiguous.
This is an important step even before a broad release. A mobile product needs a repeatable way to package, distribute, and update the client. The work in August made that path more concrete while keeping the beta boundary clear.
Liquid glass, with a stable fallback
The app’s glass treatment expanded to Android through a native adapter and a set of shared components. When the platform can provide the effect, surfaces can feel lighter and more system-aware. When it cannot, ISEA falls back to a predictable treatment instead of making the interface depend on one visual capability.
That balance matters for a cross-platform app. Visual character should travel with the product, but the underlying interaction still needs to work on different devices and OS versions. Tabs, buttons, containers, transaction sheets, and other high-traffic surfaces received compatibility and contrast refinements along the way.
Voice capture starts faster
The Agent composer now responds more immediately to voice interaction. Touch handling, press-and-hold behavior, release timing, transcript preservation, and cancellation were all tightened so the handoff from speaking to sending has fewer ambiguous moments.
Streaming replies also render as readable Markdown while they arrive. Together with the earlier waveform work, this makes the Agent feel less like a remote process and more like an active part of the app: speak naturally, see the response take shape, and keep enough context to review what happened.
Long conversations stay usable
Agent conversations can grow quickly when a request involves several records or tool steps. August added more work around long-chat responsiveness, record-list pagination, prefetching, and animation behavior. The goal is to keep scrolling, typing, and returning to history responsive even when the underlying conversation is substantial.
The Agent also gained more configurable starting prompts and clearer empty-state behavior. A blank conversation should help you begin without pretending to know more than it does; an active conversation should keep its history and actions understandable.
Sync and updates are easier to recover
Cloud sync received a complete pull action. When a user needs to restore the latest server-side state, the app has a more explicit path to request it and a more reliable way to represent the result.
The update path improved in parallel. Hot-update publishing was enabled on the server, the client can offer a restart action after an update, and remote configuration retries until the required settings are available. These changes reduce the chance that a new build or configuration leaves the app waiting in an unclear state.
This is the kind of infrastructure users notice only when it fails. In August, we kept working toward the better version: updates should arrive, sync should be recoverable, and the app should explain what it needs when it cannot continue yet.
Feedback is closer to the moment it happens
Agent messages can now carry an in-chat feedback action. If a response is confusing or a record change needs attention, reporting it does not require leaving the conversation and finding a separate form.
The feedback board and roadmap workflow were also refined so reports can be reviewed with more context. This closes a useful loop between the moment a user notices something and the work needed to improve it.
Under the surface, the Agent runtime moved to AI SDK 7, gained another compatible model-provider path, and tightened timestamp validation around record operations. Those changes give the system a more reliable base for the client-facing work above.
August was about making ISEA more reachable, more adaptable across devices, and easier to recover when the network or an update gets in the way. The direction for the next month is clear: keep the mobile surface calm while giving the Agent enough speed, context, and feedback to become genuinely useful in daily life.