October 4
What's New in QORE
We're excited to roll out this release, bringing a unified AI Gateway for every agent, an Operator that runs on a schedule, and stronger sign-in, evaluation, and knowledge management capabilities.
Here's everything that's new.
AI Gateway for agents
Every agent call goes through the AI Gateway: chat, reasoning, tool calls, document classification, summaries, conversation titles, and knowledge queries.
Model families: OpenAI and Gemini in production. Other model families are on Dev and RC only.
One place for model control: each tenant's enabled models, usage tracking, and provider error handling are applied in one place.
Picture this: A prospect asks who controls which models teams can use. The answer is one place, not one per agent.
Smarter model routing and outage recovery
The AI Gateway routes and recovers based on the required output format.
When a provider is overloaded or down, users see a plain message.
Retries no longer pile up.
Why it matters: when a provider has a bad day, users get a clear message instead of a spinning wheel.
Sign-in and access
Microsoft SSO and Entra ID support, with directory support.
RBAC revamp, with capabilities.
Authentication tab on agents: builders can require callers to sign in through Microsoft Entra or Google.
Embedded chat widgets can pass the end user's sign-in token, or users can enter it with the new Setup Identity Token button (available on request).
Picture this: An internal HR assistant only people signed in through the company's identity provider can use.
Visibility and control
Pinned projects: pin the projects you use most often, so you can resume work on them every time.
Stopping a system agent now also stops every run and helper agent it started.
New Connectors page in Control Center:
Tabs: Integrations, MCP Servers, Platform Skills.
Next release target → shows which agent uses each one, with filtering and search.
Embedding Logs in LLM Logs record every call to an embedding model.
Agent export by email
Large exports used to time out in the browser.
Now you get an email with a download link, plus any export warnings.
Operator runs and manages complex work across the platform.
Automation
Schedule triggers: run an Operator daily, hourly, and so on, with no manual start.
Picture this: A recurring task kicks off every morning before the team logs in.
Long-running builds in the background:
App Builder and Knowledge Builder tasks run as tracked background tasks, not one call that can time out.
The Operator shows progress and picks up the result when the task finishes.
Staying in control
Gates in the top bar of Console and Control Center:
Lists every conversation you started that stopped to ask a question or for permission.
Nothing waiting on you gets lost in chat history.
File uploads in Ask User: an agent can ask you to upload a file as your answer, and these show up in Gates.
Per-message model and effort choice: applies to that message only, not the whole conversation.
Connections
List integrations tool: the Operator can list available and connected integrations, and tells you which connection to switch on or connect when it lacks a tool.
Connection selection in the Operator composer and dashboard input.
More Operator capabilities
Account Management MCP: manage the platform account from the Operator.
Doc Template create, read, update, and delete support in Operator tools.
MCP connections
Connect MCP servers with OAuth client credentials (M2M), with no user sign-in needed.
One badly shaped tool no longer breaks the whole server connection.
The per-user sign-in link now shows up in chat while the tool runs.
Agent Builder and Flow Builder
Agent Builder and Flow Builder agents are now on the latest engine (executor 1.4), the same as user agents, with newer capabilities and model support.
Export App and Export Agent moved to the right-panel header.
Chats are saved in chat history, so you can start a new chat on the same agent while retaining the past conversation.
Redesigned Agent Builder header with Flow and Overview tabs.
New Report Issue action in the chat header.
New agent capabilities
The agent that builds agents now has Knowledge Builder agent skills.
Knowledge Graph Search skill for agents, to query a governed graph.
New Platform Q&A agent, backed by the GitBook documentation.
Fixes
Evaluations and Human Review help teams measure agent quality and review outcomes.
Offline evals
Build an offline eval dataset from an uploaded document, on a new in-page creation screen.
Picture this: Upload a document and provide the test cases. You can also create auto-generated test cases using AI.
Online evals
Online evals include faithfulness evaluation. It checks whether the agent's answer stays true to source documents and tool results, and flags made-up content.
Picture this: A direct answer to “How do we know the agent isn't making things up?”
Runs and scores
Per-evaluator runs with Pending and Failed states:
Each evaluator gets its own run.
See queued, running, completed, and failed evaluations, not just finished scores.
A failed run shows its reason.
New Runs tab in Control Center → Online Eval.
Project filter on Scores and Runs, so agents with the same ID in different projects no longer get mixed up.
More detail on each score:
Shows the points the judge based its verdict on.
Links to the exact conversation judged, in Online Eval and Human Review.
Assign agents to an evaluator, with clearer on/off controls per agent.
Screen polish: back button in the eval detail header, and a Scored column in the Online Eval list.
Models and billing
Judge models:
The picker lists only models the tenant has enabled in the AI Gateway.
Now searchable.
Jury models can be mixed across providers.
Billing: online-eval model usage appears on its own line in Cost Control → Credit Usage → Review credit usage.
Knowledge Hub and Knowledge Builder help teams create, manage, and use governed knowledge.
Keeping knowledge current
Upload new versions of files from Google Drive, OneDrive, SharePoint, and websites to existing datasets.
Knowledge and file versioning.
Picture this: A policy document gets updated. Add its new version to the existing dataset instead of starting over.
Finding and checking content
Semantic search on data sources, on the Dataset page and in Knowledge Builder and Agent Builder chat.
Preview before sync: summary and raw extraction results in a dataset view with intermediate results.
New Preview tab and Dataset Sync Ledger in Knowledge Hub.
Ingestion
Combine crawled website pages into a single data source.
Duplicate ingestion issue fixed (hotfix on 2.15.2).
Google Drive uploads are no longer limited to PDFs.
Experience
Redesigned Knowledge Builder header.
Content tab aligned to the new design, with table-of-contents scrolling.
Knowledge Playground moved to its own page.
Govern Graph: view, download, and rename.
Design System Builder: new agent and chat session for building design systems.
Configured fonts are applied in generated applications.
Design system versioning.
Picture this: Generated apps can follow a customer's own design system, including its fonts.
Design System home page with tables, filters, and font selection.
Model search, with the default model logo shown.
Pagination for Operator App Builder versions.
Images rendered inside Markdown messages.
Fixes:
Long-running App Builder sessions no longer hang.
AI Gateway caching and thought-signature fixes address App Builder hallucinations.
Standardized modal sizes and a same-name menu pattern across builders.
Tidier spacing and a shared help banner.
Widgets SDK: Show More on long user messages.
Dark mode on dev and RC only, for the QORE default theme.
We'll continue to expand on these capabilities in upcoming releases.
Let us know what you think of this upgrade—keep building, and keep the feedback coming!
Maintained by Abhishek Paul
AI-assisted, human-approved
Last updated
Was this helpful?