The moments worth noting, most recent first. Small enough to read in a sitting. For what's coming next, see the roadmap.
Beta opens
Tessera is open for a wider beta. The product has moved to tessera.app and the site is rebuilt around the partnership story: not "AI generates a doc," but "AI co-authors a product model with you."
Free during beta. No credit card. Beta users shape what ships next.
Partnership reset
Kickoff is now a conversation, not a one-shot generation. The AI reads your inputs deeply, surfaces tensions and pushbacks, and makes unilateral calls that invite you to disagree.
Per-question alternatives. The AI generates plausible answers for every question alongside its own lean, so confirming or steering takes one click instead of a paragraph.
New kickoff surface
Brand new welcome composer at the home screen. Single field, paperclip file attachment, recent projects, resume any in-flight kickoff in one click.
Conversation UI with question cards, numbered answer alternatives, and a Something Else inline answer for nuanced replies.
Live model outline that ticks DISCUSSED, INFERRED, and FROM YOU as you talk, so the partnership progress is visible in real time.
Preparing surface rewrite with a 3-segment progress bar, prior-answers strip, and a Discard option that cleanly cancels finalize in flight.
Provenance system
Every field on every node now carries its source. You said it directly, it came from a document, it emerged in conversation, or the AI inferred it.
Segmented provenance bar at the bottom of every card. Greens for what you authored, blues for what the AI inferred, gray for what's still open.
View sources collapsible in card details shows per-field attribution. Scan any card and see exactly which fields came from where.
Build package export
Tessera's primary output shipped. Download a structured zip and drop it into Claude Code, Cursor, or Codex CLI. AGENTS.md, PLAN.md, context sidecars, design tokens, and per-tool rules all in one package.
Copy buttons for Bolt, v0, and Lovable in the same drawer. The right artifact for each tool: blueprint prompt for Bolt, shadcn registry for v0, Knowledge Base for Lovable.
Live Readiness score and counts-based diff in the drawer. See what changed since your last export before you ship v2.
Flows and scenarios become first-class
Scenarios are first-class. Build concrete execution paths through your flows: happy path, error recovery, edge case, abandonment. Each scenario has a type, outcome, persona, goal, narrative, and a specific path through flow nodes.
Trace mode for path authoring. Click nodes in the flow editor to build a scenario path; visual feedback shows you the route as you go.
Real auto-layout for the flow editor. Click Auto-layout and complex flows reorganize themselves with topological structure intact. Branches at the same rank sit side-by-side, diamond patterns render as diamonds, complex flows read top-to-bottom without manual nudging.
Horizontal orientation toggle for flows. Switch between top-to-bottom and left-to-right per flow.
Multi-lane back-edge routing. Loop-back edges route through dedicated lanes so they don't overlap visible content.
Quality pass and Readiness score
Kickoff review pass. Every generation runs a second LLM call that critiques the model, adds entity states, scenarios, and field-level improvements. Typical brief picks up around 22 entity states, 9 scenarios per flow set, and 20+ sharpened fields.
Entity cards now show domain-appropriate states. An Appointment entity shows Scheduled, Completed, No-Show, Cancelled instead of generic placeholders.
Readiness score in the preview. A single 0 to 100 score telling you whether your model is ready to export, plus a concrete list of things to fix.