Happier Server v0.2.12-preview.1

Happier 0.2.12 focuses on keeping long-running work available and recoverable across reconnects, provider limits, daemon restarts and device changes. It also improves file and code-review workflows, terminal setup, notifications and self-hosted operation.

Sessions keep moving through interruptions

• Messages sent to inactive sessions can authorize the owning daemon to resume the session and deliver the work without requiring the app to remain connected.
• Pending messages and turn mutations are retained more reliably across reconnects, restarts and temporary delivery failures.
• Failed activation attempts remain visible and recoverable instead of silently losing the requested work.
• Session drafts, new-session drafts and review comments are preserved when account or server scopes change.
• Transcript catch-up, viewport restoration and streaming Markdown remain stable while sessions reconnect or receive additional history.
• Managed Agent launches and handoffs retain their operation state and clean up temporary transfer files after durable export.

Connected Services and provider recovery

• Account pools can recover more cleanly when a provider temporarily exhausts capacity or an account reaches its usage limit.
• Provider usage, subscription state, reset times and recovery waits are retained and shared through the existing Connected Services surfaces.
• Account switching stays attached to the active provider runtime where supported, reducing unnecessary Agent restarts.
• Authentication and relay interruptions no longer discard otherwise valid stored credentials.
• Quota checks across account groups run concurrently, improving recovery responsiveness for pools with several accounts.

Daemon, terminal and Agent reliability

• Daemon startup recovers from recycled process identifiers, inconclusive process probes and interrupted tmux session state.
• Linux session resource policy now applies consistently across supported startup modes.
• Codex sessions preserve retained thread continuity, wait for prompt steerability and recover from transient app-server database contention more reliably.
• Claude terminal dialogs and permission routing are more consistent across interactive sessions.
• Pi and ACP sessions settle interrupted or failed turns without leaving misleading in-progress state.
• Setup flows add guided terminal presentation and arrow-key navigation, with clearer relay-selection and reachability guidance.

Files, Git and code review

• File, review and source-control workspaces have clearer navigation and more stable selection state.
• Diff views add wrap-line controls and improved rendering for large or streaming changes.
• Reviews can select exact files for a commit while retaining the surrounding workspace context.
• Literal file paths and file statistics are handled more safely across Git operations.
• Filesystem browsing, editing and navigation remain stable when switching between workspace surfaces.

Notifications and interface polish

• Agent requests can produce richer notification summaries while respecting the request-notification preference.
• Push delivery retries only failures that can safely be retried.
• Startup notification handoff is more resilient when the app is opened from a notification.
• Archived sessions stay out of active attention surfaces.
• Navigation, segmented controls, anchored tooltips and the Android composer have received additional stability and presentation fixes.

Self-hosting and delivery

• Stack targets retain their own synchronization state and keep remote servers behind the target-owned proxy.
• Relay uninstall and upgrade paths preserve persistent user data.
• Website download counts and fallback behavior are more reliable.
• Release publication now supports a coordinated preview-and-stable fast path while still building and verifying each channel's distinct policy-bound artifacts.

Compatibility

This release adds optional protocol fields and nullable server database columns. Existing clients continue to use their established behavior, while newer clients activate daemon-owned inactive-session recovery only when both the server and daemon advertise support. No coordinated app, CLI, daemon or relay rollout is required.