Fuel iX Copilots — Week 28

This was a full week across the board — new features landed alongside a solid round of fixes that address real day-to-day pain points, from access issues on large teams to smoother agent behavior and better visibility into usage and costs.

Delete Prisms directly from the dashboard: Prisms can now be removed from the dashboard using the 3-dot menu on each card. A confirmation prompt prevents accidental deletions, and any Prism referenced in an existing chat gracefully shows a "Prism deleted" indicator rather than breaking the conversation.
Copilot details accessible from My Copilot Hub: Users can now select "About" from a copilot card's context menu to view its details without leaving the hub, making it easier to review copilot information at a glance.
Jump-to-prompt navigation in chat: A floating up-arrow button now appears when earlier messages scroll out of view, letting you instantly jump back to your most recent off-screen prompt without manual scrolling.
Login page no longer hangs for sessions older than 24 hours: Fixed an issue where an expired authentication session could leave users stuck on a stalled login page. The flow now correctly detects the expired session and prompts you to re-authenticate right away.
Fixed access loss for members of large teams: Users on teams with 700 or more members were silently losing access to shared copilots due to an out-of-memory error during team sync. This has been fixed, and affected users' permissions have been restored.
Agents now finish gracefully when reaching their execution limit: Instead of surfacing a raw error when hitting the tool execution limit, agents now wrap up their current work and deliver a clear summary of everything gathered so far.
Parallel tool calls in agents now complete reliably: Fixed an issue where agents issuing multiple tool calls in a single turn would end up in a failed state after submitting their outputs.
Smarter knowledge base file lookup reduces unnecessary processing: Agents now look up knowledge base files only when a request actually needs them, rather than loading all file names into every prompt. This reduces overhead and keeps interactions more efficient.
Improved reliability when accessing knowledge base files: Knowledge base integration files now download through a secure proxy, improving reliability when accessing content in production environments.
Large knowledge base bulk uploads now work reliably: Bulk uploads with thousands of files (27,000+) no longer fail due to document size limits, and timeout errors now report accurately instead of showing misleading messages.
Smarter document chunking for knowledge bases: Two new options give administrators more control over how documents are split for retrieval. A new "LLM-semantic" chunking strategy uses context-aware boundaries for better retrieval quality on complex documents. Additionally, custom chunk size and overlap can now be set directly on a knowledge base, so every file uploaded automatically inherits those settings without needing per-upload configuration.
Duplicate emails skipped automatically in bulk user uploads: When uploading multiple users at once, duplicate email addresses are automatically detected and skipped, with only the first occurrence kept. A summary of all skipped duplicates is included in the upload results so administrators have full visibility.
Fixed data loss during folder sync in shared copilots: Fixed an issue where a sync triggered by one user could accidentally delete folders belonging to other users when permissions were insufficient. Folder sync now correctly tells the difference between "folder not found" and "permission denied," keeping shared content intact for everyone.
Token revocation for MCP server connections: When an MCP connection token is revoked — whether manually, on session expiry, or after a refresh failure — credentials are immediately removed and you receive an email notification, giving teams clearer visibility and control over active integrations.
More reliable tool availability across MCP sessions: MCP tool caching now stores data at the user collection level, delivering more consistent tool availability from session to session.
Clearer OAuth permission guidance when connecting integrations: When connecting integrations such as Atlassian or GitHub, the authorization screen now shows the supported permission scopes available from the connected service, making it easier to understand what access you're granting.
Manager usage reports now reflect the full organizational hierarchy: Managers can now view usage data across their entire reporting chain — not just direct reports — ensuring accurate visibility at every level of the organization.
Per-user usage breakdown for managers: Managers can now view AI usage and costs broken down by individual team members, making it easier to identify high-cost users and allocate resources effectively.
Service account usage visible in Analytics: Developers can now view combined usage and costs for personal projects and owned service accounts in a single unified view. A new Type column distinguishes between "Personal" and "Service account" entries, and a grouped project filter makes it easy to slice data by project.
Improved project filter accuracy in Dev Portal Analytics: The project filter now shows correct checkbox states, displays "All Projects" when everything is selected, and includes an "Only" quick-filter per project. Service accounts with zero usage are also listed reliably, so no projects go missing from your analytics view.
Clearer guidance when creating service accounts: The warning banner in the Create Service Account flow now clearly explains that service accounts are intended for mature, multi-user use cases, helping teams decide when to use a service account versus personal credentials.