Highlights
- Breaking: Planning Mode is now an open-ended interview you explicitly validate — no more fixed depth caps or checkpoint gating
- Security: in-app reports scrub the full activity trace before filing, so paths and tokens never leave your machine
- Pick your update track — beta or stable release channels in Settings or
fn update --channel <stable|beta> - Your custom workflow now fully drives the board — cards move through the columns you actually defined, not a fixed six
- Review artifacts: auto-generated task verification videos and deliverable galleries in a new Quality hub
New
- Beta and stable release channels — choose your update track in Settings or
fn update --channel <stable|beta>. - Your workflow now drives the board — cards move through the columns you define, not a fixed six-column set.
- Review artifact controls: auto-generated task verification videos and deliverable galleries in a new Quality hub.
- Guided in-app Bug/Feedback/Idea/Help reporting with screenshots, GitHub Issues or Discussions filing, and roadmap deduplication.
- Mission auto-merge override — bundle a mission's features onto one branch and one PR.
- Mailbox approval for ephemeral agent follow-up tasks.
- Ideation: persisted sessions with promotion of research findings into mission roadmap features.
- Mission hierarchy tools exposed to agents and dashboard chat; scheduled mission work with symbol-level concurrency control.
- Portable org export/import (
fn org-export/fn org-import), now surfaced in the Team tab. - Configuration revision history and rollback, now in Settings.
- Native structure previews (missions, findings, evals, goals, roadmap items) in chat and Mail, with drag-to-attach.
- Optional Task chat progress feed narrating steps, failures, reviews, and rollbacks.
- Dashboard chat agents can now edit files and run bash directly with coding workspace tools.
- WhatsApp plugin: pairing QR code and setup instructions surfaced in plugin settings.
- Embedded PostgreSQL connection cap is now configurable in Advanced Settings.
- Planning Mode simplified into a sequential flow: question → plan review → refine/validate.
Fixed
- Planning Mode: dozens of reliability fixes — stable timers and session restore across refresh/stop/resume, reliable refinement submission on mobile, synced questions and running plan, durable initial plan generation, mobile/tablet layout, collapsed history by default, and rejection of truncated final plans.
- SQLite→PostgreSQL migration status now surfaces on the dashboard while cutover is pending; several PostgreSQL stability fixes (permission errors reading migration health, concurrent project-init races, NUL-byte crashes in chat/mailbox writes, restored activity logs).
- Workflow/board fixes: no-merge workflows now finish in their completion column, custom Merging columns receive cards correctly, hold nodes no longer crash the Pull Request workflow, and tasks without a saved workflow selection can move again.
- Fixed Codex weekly usage reporting, Grok/Claude MCP bridge packaging, OMP (Oh My Pi) model runtime failures, and GitHub Copilot re-login banner persistence.
- Task detail and Kanban UI polish: consistent action-button sizing across themes/breakpoints, mobile swipe settling on one column, cost badge placement, and deduplicated agent name badges.
- Restored translations (zh-CN/zh-TW roadmap duplicate label; Settings Configuration Versions across five locales).
- Duplicate-task guards: prevented duplicate follow-ups from retried steps, cross-parent diagnostic duplicates, and stale "needs your decision" states.
- GitHub/GitLab: fixed foreign-language issue-form auto-translation and forced transport selection for Discussions.
- Fixed a dashboard build failure caused by a missing dependency; Settings now autosaves instead of requiring an explicit Save.
- Windows update timeouts extended; Compound Engineering personas restored in npm installs; Claude CLI pi extension load fixed via SDK dependency pinning.
Breaking
- Removed the Planning Mode deepening checkpoint and fixed interview depth caps — plans are now validated explicitly by the operator instead of by AI completion.
Security
- In-app report filing now scrubs the full activity trace before it leaves your machine, so paths and tokens never reach the pipeline.
Internal
- Review gates (plan/code/browser) now run exclusively as workflow-graph nodes; the legacy in-session step reviewer and its git-reset rewind path are removed.
- Tasks left mid-flight by an older Fusion version are now adopted cleanly on upgrade instead of getting stuck.