The latest releases and improvements to Proto Studio.
The new Inferences API lets you call a model directly, without setting up an agent or a conversation. Generate text, create and edit images, turn text into speech, transcribe audio, create embeddings, and get structured decisions, all through one endpoint. Get results right away, stream them as they are generated, or queue longer jobs and pick them up later.
Try a model before you build with it. The new Playground in the AI tool lets you pick a model, fill in a prompt, and see the result in seconds, with ready-made examples to get you started. It works with every kind of model, from text and images to speech, transcription, and decisions. You can even record audio straight from your microphone to test transcription.
Speech to text now streams. Transcripts appear as people speak instead of after the audio ends, so you can build live captions, voice notes, and voice-driven apps. Each line includes timestamps and, when the model can tell, who is speaking. Export finished transcripts as text, JSON, or SRT and WebVTT subtitles.
More OpenAI models are now available in the model catalog. Enable them for your organization and use them in your agents, conversations, and the new Inferences API.
Speech to text, text to speech, and decision models are listed under AI on the pricing page.
Every model field in the AI tool and on the canvas is now searchable, showing each model's provider and type and pinning recently used models. Browse opens the full catalog with type filters and details, and can enable a model and select it in one step.
Pricing for time series databases has been updated and we corrected a pricing error where older versions of some LLM models were being charged frontier rates. See the pricing page.
A new organization Usage page shows spending for the last 30 days and the quotas for the current billing term. It includes a total, how much of that spend is included in the plan, a daily chart by service, a breakdown by resource and rate, and quantity limits. Usage totals are updated at least every five minutes.
The time series API has been improved to include aggregates and grouping, and the management area now lets you browser your time series database tables using a customizable chart and table.
A new static check tools runs automatically after AI changes are made. It checks the project, code, and design for common errors. It is a cheaper and faster way to catch and fix errors than the more comprehensive AI QA agent. You can still run the QA agent like before release.
The Studio editor has a major upgrade. More than a dozen bugs are fixed. Agents are more efficient, and QA catches issues earlier. You have more control over page design, and clearer visual control over loops and conditionals.
You can add an OpenAI, Claude, or LM Studio model for an organization and use it with agents and conversations. OpenAI and Claude models use an enabled, ready password or OAuth connection.
Package pages on proto.zip again include the Go module metadata in the page head, so Go can resolve proto.zip/studio packages.
Open a gateway and choose Watch logs to see recent HTTP requests and follow new ones as they finish. Each line includes the status, method, path, host, and duration.
TypeSafe AI, Inc. (Jev) is now a subprocessor for machine learning services. The updated list is on the subprocessors page, which the Data Processing Addendum incorporates.
Fixed a bug, present since the first day of beta, that occurred when the Studio editor was left open for more than an hour.
Cloud, edge, and HTTP runtime functions have greater log observability. Open a function or deployment and choose Watch logs to see recent lines and follow new ones.
Session-scoped refresh tokens are returned by default with online sessions, not only with offline tokens.
Security settings now include additional control over token lifetime, and a global token invalidation list is available via API.
Real-time updates in the project canvas are better.
Custom CSS support is improved. CSS and font edits remeasure the artboard once the new styles are in the document.
Several page editor bugs are fixed. Auto-height artboards size to their content, and an empty frame no longer sticks at full viewport height.
Saving a project canvas is about 10–20× faster.
Real-time event stream subscriptions have additional features. The canvas shows Reconnecting when the resource stream drops.
Project canvases now sync across windows and sessions. This release also includes several bug fixes and performance improvements.
Auto-layout on AI-generated projects is improved.
AI tool use is about 3–8× faster.
The project API is about 3–8× faster.
You can remember specific devices to bypass MFA on future logins from the same device.
All users can now create and manage organization profiles.
Agents are more efficient and have higher capacity limits.
All users can now update their profile and security settings.
Anyone can now sign up for the free plan and join the Proto Studio open beta.