Release overview — July 2025
The Curiosity Team
July gave the assistant tools: it can now call them while it answers, Curiosity Workspace can offer them to other AI applications over MCP, and new workspaces ship with a few ready to use. Four releases this month.
An assistant that takes actions
- Tool calling: the chat assistant can call tools while it answers, so it can take actions and pull in live data instead of only generating text. The conversation shows which tools were invoked.
- Built-in tools out of the box: new workspaces ship with a document search tool and an image generation tool. Tools can store generated files and attach result snippets that surface back to the user.
- Richer tool authoring: one tool definition can expose several callable functions, and parameters can declare default values that apply when the model omits them. Custom tools are validated before they run.
- Translate files with AI, with error messages that say which part of the file failed and why (moderation, token limit or connection).
"Your workspace as an MCP server"
A workspace can now host its tools over the Model Context Protocol, so external MCP clients can discover and call them. That makes Curiosity Workspace a tool provider for other AI applications.
Results inside the answer
The assistant can show source documents and generated files inline in a chat message, as compact cards. Clicking a card opens the item in the side preview; right-clicking it brings up the same command menu as on search results.
Filters that stay put
- Pin, lock and hide filters: pin your most-used filters to the top, lock a filter so its selection carries across searches, and hide the ones you don't need. The choices are remembered between sessions.
- Per-area filters: each search area can show its own set of filters.
- Search can exclude specific node types from result merging, and embedders can cap result height and post-process facets before they render.
Sign in your way
- Map external sign-ins: a custom authentication flow reads identity details (email, username, name, admin status, department, company) from a bearer token and creates or updates the matching workspace user, for setups behind external identity providers and reverse proxies.
- Extra diagnostic logging for that flow can be turned on at startup.
- The login flow reuses an existing valid session before asking for credentials, and new workspaces get stronger session token signing keys.
For developers and operators
- Git repository sync from the CLI pushes repository state to a workspace through the git tracker. Ctrl+C now stops a served front end cleanly.
- Custom code endpoints relay progress to their callers by default; the
progress call is now named
RelayStatusAsync. - Availability, deployment identity and health endpoints sit at a fixed root location whatever the configured base path.
- Lower memory use for very large graphs, and faster vector search from an embeddings index optimized on save.
The little things
- Reconnecting after a dropped connection refreshes the chat so no messages are lost, and duplicate replies no longer appear.
- Past chats stay listed even after their AI provider is removed.
- Confluence sync keeps a cleaner folder hierarchy, tracks update checks per user, and flags connections that need re-authentication.
- Amazon Bedrock tool calls use consistent identifiers across multi-step interactions.
This overview covers builds v25.7.59609 → v25.8.60114. For the full, build-by-build detail see the changelog.
Read next
Articles on context graphs, enterprise search and industrial AI