// Changelog
New features, changes, and fixes shipped to TarqaAI.
Build tools with a form instead of hand-writing JSON (the JSON editor is still there for advanced cases). Tool calls now show a readable card — the tool name in plain English and its arguments as a labelled list, with raw JSON one click away — and the result box explains what to do. Tool calls are kept when you reopen a conversation, and tool calling now works with your own GPU-tunnel (hosted) models too.
Two-factor sign-in now uses a one-time code sent to your email. The authenticator app (TOTP) and passkey options have been removed.
Account, billing, and team emails have a refreshed, consistent layout.
Fixed duplicate payment handling, subscription cancel/retry, and cross-team isolation in Razorpay webhook processing.
Improved document chunking and text extraction, and fixed memory extraction pulling in irrelevant context.
Plan filtering, pagination, CSV export, and an expanded audit log across user and team management.
Fixed webhook delivery retries and duplicate project name handling in team workspaces.
Closed fail-open paths when the MFA encryption key is misconfigured, stopped malformed verification requests from returning a 500, and fixed a shared rate-limit lockout during verification.
Connect Google Drive or Notion to sync documents straight into your knowledge base, with automatic background re-syncing.
Verify sign-in with a one-time code sent to your email as an alternative second factor.
Added CAPTCHA and disposable-email checks on signup and login to block automated abuse.
The chat API now supports tool calling across Claude, GPT, Gemini, and Bedrock models — the model can request a call into your own application (e.g. to query your database) and use the result to answer.
Register your functions once and the JS/Python SDKs auto-discover them on every call — no need to pass tool schemas and handlers each time. You can also let the model search your knowledge base as a tool it calls itself.
Fixed a CORS configuration issue that could block legitimate cross-origin requests to the API.
Code and document responses in chat now render as downloadable artifact cards with correct file extensions.
Resolved several high-severity security and billing issues found in an internal audit, and removed unused legacy code paths.
You can now subscribe to product updates via a newsletter signup.
Fixed inaccurate token and usage counts shown in analytics, billing, and the admin panel.
Securely reveal a previously created API key (with MFA verification) instead of having to generate a new one.
Fixed a billing calculation issue affecting team plan subscriptions.
Fixed the docs site crashing after deploys and serving stale content for up to a year on repeat visits.
Improved admin tools for managing team seats and billing subscriptions.
Fixed the search scope dropdown not finding indexed documents, and RAG chat answers pulling in unrelated knowledge base content instead of the documents attached to the conversation.
Added a Contact Us form for reaching the team directly, and this changelog page to track product updates going forward.
Team workspaces now automatically downgrade when a subscription lapses, keeping billing and access in sync.
Improved search and live monitoring tools in the admin panel for faster account and usage lookups.
Reorganized docs with new guides for GPU tunneling and hosted model deployment.
Billing now tracks usage per workspace, giving teams clearer visibility into spend across projects.
Chat responses and documentation now render Mermaid diagrams inline.
Added Vertex AI as a model provider, plus embeddings support for retrieval-augmented chat.
Team workspaces now have an activity log tracking member actions, API key usage, and knowledge base changes.