May 2026Release log31 May 2026
This month we replaced ad-hoc AI generation with a unified René Engine, moved generation and translation onto a background job queue, and fixed a long list of workspace ownership, billing, and scoring issues.
Generation engine
6 changes- Journey creation now runs on the René Engine instead of direct AI calls, giving more consistent intent classification and stage sequencing. This includes a learned sequence model and data-driven stage ordering so mid-purchase stages rank more realistically.
- Added archetype coherence checks and canonical stage filtering so generated journeys stay aligned to a chosen archetype rather than drifting between templates.
- Stage selection now takes focus keywords and industry resolution into account, and user design directives are respected during generation.
- Rebuilt the AI core into a shared runner used across journey, archetype, ritual, VOC, and behavioral-plan generation, with routes and dialogs migrated to a queue-and-hydration model so long-running jobs no longer block the UI.
- Fixed AI hallucination of short numeric IDs, stuck-flag bugs, and hangs in LLM calls; added a self-healing reaper that recovers stuck generation jobs automatically.
- Added a fourth behavioral-plan generator, editable ritual cards, and a review mode for archetypes with a new dialog.
Translation and canvas
4 changes- Journey translation moved to background jobs with chunking, a rehydrator, and parallelised stages, and migrated from Claude to Gemini and then to the Vercel AI Gateway for translation calls.
- Added a flat-text fallback for RTL languages and for translation failures, fixed translation progress display to update monotonically, and resolved exhaustiveness and language-label issues.
- Fixed canvas transformer scoring drift and improved canvas rendering, dock overflow, and z-index issues.
- Redesigned the generation dock into a single panel and added a spotlight detector to help locate jobs and content.
Scoring
4 changes- Implemented a scoring engine with analytics integration and routed all score reads through a single calculation function for consistency.
- Refined touchpoint scoring and the gradient bar visualisation, and fixed a hook-count crash in the touchpoint panel.
- Renamed the scoring terminology to EXIS across the UI and documentation, fixed a related dashboard calculation bug, and added a scoring FAQ page to the admin nav.
- Added a theme-aware Customer Emotional Arc strip.
Workspace and billing
6 changes- Fixed multiple workspace ownership transfer bugs: incorrect data propagation, permission checks, and missing immediate feedback after a transfer.
- Restricted the Transfer and Danger Zone cards to admin/owner roles only, and resolved workspace delete and contributor role issues.
- Completed the subscription cancellation flow with state-aware controls, split credit pools with a UI update, and normalized billing cycle values to "yearly" at the API boundary.
- Added Enterprise+ tier support: admin sidebar entry, super-admin provisioning, in-house invoicing, and expanded plan_id constraints.
- Implemented a plan expiry cron job, PDF invoice generation, and updated Stripe webhook and cron handling for the plan-tier refresh.
- Corrected Gemini billing rates across endpoints and clarified AI Gateway model references, including an upgrade to Sonnet 4.6.