What happened
Mika’s GitHub release notes identify v0.1.45 as a release published on November 16. The update combines a rollback of voice wake-up, changes to scheduled-task controls, and a range of bug fixes and usability adjustments. The voice-wake rollback is the clearest product decision in the release: the maintainers say an ordinary app without root access cannot reach Android’s system hotword entry point. Keeping the microphone open continuously at the application layer does not solve the problem, because power consumption and background-execution restrictions remain. The notes say the existing press-and-hold voice features, including long-press-to-speak from the capsule interface, are unaffected.
For scheduled tasks, the release adds a setting for maximum runtime. It also changes the time-planning interface: users can edit multiple times in daily or weekly schedules through a time picker. The notes do not specify a default runtime, the available range of runtime limits, or how a task behaves when it reaches its configured maximum. They do not describe whether the time picker changes how existing schedules are stored or migrated. These are changes to configuring and bounding recurring work, rather than a claim that Mika can execute a new category of task.
Several fixes address error handling and recovery. Mika now distinguishes a provider response of 403 with insufficient_quota from an invalid API key, reporting that the quota is exhausted. Daily-budget fallback selection now chooses a backup model by provider ID, so exhausted quota should no longer send the user to another model at the same provider. The release also fixes a mistaken microphone-busy warning when a recognition service fails to start, adds guidance when that service lacks recording permission, and notes compatibility with HyperOS’s default recognition entry point.
The remaining changes include validation for API keys containing invalid characters such as newlines, Chinese characters, or emoji; a clear button for that configuration; and a fix for a phone that remained at an “unlock your phone” wait state after the screen was on but unlocked. Other listed adjustments cover copying selected text from assistant replies, preserving the message snapshot selected for session export, clearer failure notices for personal skills and Feishu locks, and more useful ADB error reporting if a log export disconnects. The release notes also mention editable WeChat follow-up drafts after AI calls and interface adjustments to the conversation sidebar.
What the archive adds
New Runtime’s earlier coverage on Mika v0.1.13 (October 31, 2026) described USB startup, interaction pacing controls, and OpenAI Responses API providers. This is a new event in the same product’s development, not a restatement: v0.1.45 documents a voice-wake rollback and new scheduling controls.
Trend signal
This does not yet fit a pattern New Runtime tracks. Mika’s decision to remove a voice-wake feature because Android’s system access and background limits constrain an ordinary app is a concrete platform trade-off, but this release alone does not establish a broader runtime-boundary trend.
Connections to other events
Mika’s earlier v0.1.13 release added USB-based startup and interaction pacing controls, as New Runtime covered on October 31, 2026. The current release continues the same project’s work on an Android assistant, but shifts from adding capabilities to correcting a voice interaction that could not be made system-level under the stated app constraints. The new scheduled-task limits are a separate operational change.
Why it matters
A voice interface that appears always available is not necessarily feasible as an ordinary Android app. Mika’s rollback makes the trade-off explicit: application-level continuous recording cannot substitute for system hotword access when power use and background restrictions apply. For teams building mobile assistants, that distinction matters before investing in a wake-word experience that depends on privileges their app does not have.
The scheduling changes address a different reliability question: how long a recurring task may run and when it should run. A maximum runtime can make scheduled work more bounded, while editable multiple times can support schedules with more than one daily or weekly run. The release does not establish how these controls behave in practice, but their presence points to configuration and execution limits as part of making assistant workflows manageable—not just adding new ways to invoke them.
Limits and uncertainty
The release notes explain the rationale for removing voice wake-up, but do not provide measurements of battery use, background reliability, or user impact. They do not say whether the feature might return through a privileged system integration or another implementation. For scheduling, the notes give no maximum-runtime defaults or bounds, and do not describe what happens to a task when the limit is reached. The release also provides no comparative evaluation of the fixes or the reliability gains they produce.
Skill to build now
When designing an Android assistant, separate system-level wake-up from in-app voice controls at the product-planning stage. If an app lacks system hotword access, offer an explicit interaction such as press-and-hold rather than relying on continuous app-level listening. For recurring agent work, define a maximum runtime and test schedules with multiple daily or weekly times before treating the workflow as dependable.
Reader takeaway
Mika v0.1.45 is notable both for what it removes and what it bounds: system-style voice wake proved incompatible with the stated constraints of an ordinary Android app, while scheduled tasks gained runtime and timing controls. For mobile agent builders, access to the operating system is a product constraint, not something an always-listening loop can simply work around.