Scalar Agent release notes
Polish and bug fixes
API key prefixes and cleaner chat context
- Header API keys now support token prefixes via x-scalar-token-prefix, automatically adding prefixes like "Bearer" or "token" to outgoing requests while keeping stored credentials clean.
- Selected API context is now sent as structured metadata rather than appended to user messages, keeping your questions unchanged while ensuring the backend knows which API and version to use.
Polish and bug fixes shipped
Plugin document types and TypeScript 6 support
This release adds support for custom document types through plugins and upgrades the TypeScript compiler to version 6.
- Plugins can now register custom document types alongside OpenAPI and AsyncAPI, with their own detection, traversal, and rendering logic.
- Plugin documents are kept separate from OpenAPI and AsyncAPI processing, preventing unsupported operations like Ask AI on plugin-specific formats.
- Document classification remains consistent when replacing, hydrating, or rebasing plugin documents in the workspace.
- The TypeScript compiler has been upgraded to version 6, preserving package output paths and application aliases during builds.
Improved Vite compatibility and cleaner dependencies
Polish and bug fixes
Require approval for all API requests and redact credentials
Agent Chat now requires explicit approval before executing any AI-generated API request, including GET requests. Saved credentials are validated and redacted from response data before being sent to the chat model.
- All API requests now require approval before execution, preventing automatic data access with stored credentials.
- Request destinations are validated before attaching credentials to prevent sending them to model-supplied URLs.
- Saved credentials are redacted from response URLs, headers, and bodies before sending results to the chat backend.
- The approval preview shows the resolved API destination and method before the request is sent.
- Requests are rejected when no usable server is configured, prompting API authors to add a server to their description.
Security hardening and agent key document fix
This release fixes a security issue in the CORS proxy and restores full-mode chat functionality while backend support is being finalized.
- The CORS proxy now protects against XSS, open redirect, and origin abuse attacks by rejecting document requests and sandboxing responses.
- Response headers in post-response scripts, agent logs, and request history now show the values your API sent, not the proxy's security headers.
- Full-mode Ask AI uses the agent key's registered documents again, fixing questions that require API context.
- Preview mode continues to upload the selected API description as before.
Polish and bug fixes shipped
AsyncAPI support in Ask AI
Ask AI now works with both OpenAPI and AsyncAPI descriptions. The chat interface understands AsyncAPI channels, operations, and messages, and switching between different API sources starts a fresh conversation with the correct context.
- Ask AI now supports AsyncAPI descriptions alongside OpenAPI
- Search results display AsyncAPI channels, operations, and messages with their native identities
- Switching between API sources preserves the selected description and starts a new conversation
- Unsupported AsyncAPI execution actions return a helpful explanation instead of attempting HTTP requests
Security patch and code quality improvements
This release addresses a security advisory in the AI SDK response handling and cleans up code quality issues across the package.
- Updated AI SDK dependencies to patch GHSA-866g-f22w-33x8 in response body handling
- Cleaned up lint warnings while preserving all existing runtime behavior
Polish and bug fixes shipped
Faster startup and smarter OpenAPI reference resolution
This release makes the agent chat load faster by deferring the API client modal until you actually need it. We also improved how relative references resolve in OpenAPI 3.2 documents, honoring the $self identifier and preserving authored reference spellings.
- The API client modal now loads on demand when you open a request, saving 292 KB (28%) on initial page load.
- Added support for OpenAPI 3.2 $self to resolve relative references against the document's declared identity instead of its network location.
- Preserved authored reference spellings through partial bundles and editable exports, so schema labels and anchors keep their existing behavior.
- URI resolution now correctly handles root-relative URLs, protocol-relative URLs, query strings, and fragments across all bundler workflows.
Security and infrastructure improvements
Polish and bug fixes shipped
Better international input and NDJSON response rendering
- Fixed Enter key sending messages too early when typing Japanese, Chinese, or Korean text on macOS Chrome and Safari.
- NDJSON responses now render as pretty-printed JSON records instead of showing as a binary file.
- Removed a debug log statement that fired every time you toggled the response view.
Updated package README with refreshed platform overview
Border radius now scales from a single variable
Setting --scalar-radius now rescales every corner in the interface, including larger radii and pills. Previously, overriding the base radius left some corners rounded because tokens were independent.
- All border radius tokens now derive from --scalar-radius, so setting it to 0 squares off the entire interface
- Two new tokens, --scalar-radius-2xl (12px) and --scalar-radius-3xl (16px), fill out the scale
- The rounded-2xl and rounded-3xl Tailwind utilities now work correctly
- Default values are unchanged, so existing themes render identically unless you were overriding --scalar-radius alone
Remove duplicate heading from start panel
The Agent Chat start panel no longer renders a hidden heading tag, preventing SEO issues on API reference pages where crawlers were detecting duplicate headings.
Polish and bug fixes shipped
AsyncAPI documents in the Agent workspace
The Agent chat UI now understands workspace documents that can be either OpenAPI or AsyncAPI, so mixed API descriptions load through the same flow.
Smaller initial bundle for the chat UI
Tailwind CSS is now code-split so the Agent chat interface loads less CSS up front.
More reliable document loading from the platform
Document fetches now include the platform token so authenticated doc requests succeed consistently.
Clearer messaging around Agent credits
We updated the copy around Agent credits so free limits and usage are easier to understand in the chat UI.
What did you think of this content?