June changelog

This release is centered on account clarity. Billing now reads like a product surface instead of a raw status dump, with a clearer plan header, a more legible usage meter, and labels that better explain what free and paid users are seeing.

The work also fixes a subtle but important trust issue: subscribed accounts are treated more consistently as paid plans even when subscription status data is incomplete, which reduces confusing account-state mismatches.

Outside billing, June smooths a few visual edges by disabling native context menus in production and refining logo gradients so the desktop app feels more intentional and less noisy.

This was a focused release window: recording processing, model capability feedback, metering error handling, and production DMG notarization were the meaningful user-facing changes.

The image fallback work prevents a dead-end composer state by telling users why the current model cannot handle an attached image and guiding them to switch models.

The audio permission refresh fix keeps Settings and recording flows closer to the real macOS permission state after a user grants or changes access.

This was a small recovery-focused release, not a feature launch. The useful product work is concentrated in session restore, subagent scrolling, and account-state recovery.

The release also included the final Scribe-to-June surface rename, but the public notes keep the emphasis on behavior users notice.

The bulk of this release is security depth around agent web fetches: domain resolution, public-domain preflight, canonical URL forwarding, and rejection rules for targets that could expose local or private network resources.

A smaller desktop fix keeps the command palette visually above tab chrome, which matters when navigating June from the keyboard.

This was the deepest release window in late June, with agent controls, private web tools, note processing, billing recovery, and dictation reliability all moving together.

Agent conversations gained practical controls: slash commands, clearer tool activity rows, copy/edit actions, model-selection fixes, persisted drafts, and better handling while tools are running.

Voice and account fixes were not incidental: June improved long-dictation fallback, push-to-talk recovery, accessibility permission refresh, note processing progress, unknown funding states, subscription checkout routing, and visible billing recovery in Settings.

This release is mostly reliability work for the recording pipeline. The important user outcome is fewer dead recordings and cleaner recovery after interruptions.

Bug submission also stays accepted when platform delivery fails, so reporting a problem does not look broken because a downstream delivery path had trouble.

This was a narrow bug-fix release. Its main value is containment: a note failure should not make the rest of the app feel failed.

Issue report routing was corrected so product feedback and bug reports land in the expected project destination.

This release gives users a direct way to check for updates instead of waiting for a background update prompt.

The issue-report and messaging fixes reduce two high-friction dead ends: reports not reaching the right place, and the messaging platform hanging while loading.

This is a workflow release: users can watch meeting transcription progress live, edit agent skills, and see running routines in context.

The supporting fixes improve note retry states, dialogs behind tabs, referral overflow, tab chrome clipping, auth callbacks, and file filtering.

This release makes June feel more like a desktop workspace: tabs, draggable tab chrome, a clearer Sessions nav, and update prompts that stay contained.

Voice work focused on cleanup and speed: dictation filler-word filtering, faster paste, scoped processing errors, and protection against source labels leaking into generated notes.

The user-facing work here is small but concrete: June tries harder to restore sessions on startup and exposes Settings through native macOS entry points.

Custom local model selection shipped in this window but was later removed, so it is described historically rather than as a current stable capability.

This release is mostly distribution and update quality. Universal DMGs reduce confusion for Mac users and make the download path easier to explain.

The agent also gained better sandbox-context awareness, and June started hiding raw backend parse errors that should not be shown directly to users.

This was a compact workflow release. Routines became manageable in the app, and stopping a running agent no longer waited on the interrupt RPC before the UI responded.

The HUD transition fix keeps dictation controls from feeling jumpy during voice workflows.

The release tightens everyday navigation: clearer model controls, better empty states, full model descriptions on hover, and more polished menu bar and HUD behavior.

The system-audio onboarding prompt makes meeting recording setup more explicit before users expect June to capture call audio.

The recording fix matters for longer meetings: June no longer has to re-process the entire recording every time a new meeting turn arrives.

Pinned sessions and the menu bar mark make the app easier to return to during repeated agent workflows.

This was a broad quality release. The most important security-facing change is that June no longer needs Input Monitoring for the shortcut path.

Agent workflows gained better recovery: context overflow is explained in terms the agent can act on, CLI access can be requested in chat, and users can see or change the active model from the session bar.

Routines also became clearer, with run history, sandbox defaults, and unrestricted opt-in behavior made visible.

The session-continuity fix protects active agent work while the workspace UI remounts or refreshes.

Trial checkout preparation was made faster and safer by pre-minting checkout sessions, serializing prepare calls, and clearing prepared checkouts when identity changes.

The visible polish includes smoother session transitions, updated sidebar states, a suggested model picker, and the agent HUD.

Bundling Hermes makes the agent runtime part of the app instead of a separate moving piece users need to reason about.

Runtime isolation tightened around Unrestricted mode, deleted sessions, and side-by-side sandboxed and unrestricted operation.

Issue reports can now include diagnostics and uploaded attachments, making reports more actionable without asking users to manually gather context.

This release gives the agent a safer default boundary while preserving an explicit Full mode for sessions that need broader access.

The trial flow became more app-native: checkout sessions are minted directly from June, deep links return users to the app, and re-auth retry replaces brittle portal fallbacks.

Starter shortcuts, privacy badges, verify-page copy, and routine schedule labels were updated so users can better understand what mode June is running in.

This was a narrow account-access release. It aligned the app gate with the actual trial and subscription state users should have before entering the product.

Most of the release-window commits were review follow-ups around this billing gate rather than new interface surface area.

This was the largest early release window, covering the June rename, agent workspace polish, meeting detection, dictation controls, onboarding, settings, and verification.

Meeting work included app detection for Zoom, Teams, and Safari meeting contexts, a prompt timeout, a redesigned meeting HUD, and fixes for stuck or mispositioned meeting UI.

Agent, voice, and settings work landed together: a menu bar status, better session titles, artifact file cards, approval explanations, preserved sessions, language options, shortcut persistence, microphone playback testing, and recording-consent timing fixes.

This release established the first practical agent surface: chat persistence, markdown rendering, activity status, approval routing, and a filesystem view scoped to agent state.

Hermes moved from an external idea into an app-managed runtime that starts on launch, completes tasks after replies, and keeps session timestamps and messages reconciled.

Recording reliability also improved with stuck-recording restart fixes, helper accessibility prompts for dictation paste, and long dual-source recording stability.

This release made the early Mac app easier to ship and keep current: production DMG workflow, auto-updates, signed helper fixes, updater metadata fixes, and audio entitlements.

Voice work focused on the fundamentals: recording queue reliability, incremental transcription, source-id fixes, waveform tuning, and UI cleanup around the recorder.