Fuel iX Copilots — Week 32

This week brought a solid mix of new capabilities and reliability improvements. Highlights include better visibility into conversation capacity, expanded analytics filtering, new file format support, and a handful of fixes that make agents and sessions behave more predictably.

Filter analytics by team member: Managers can now narrow down KPI cards, usage charts, and model breakdowns to up to 25 specific team members using a searchable, multi-select filter in the analytics dashboard header.
Granular API portal access control for admins: Organization admins can now grant or revoke developer portal access for individual users directly from the Manage Users settings page, with a simple menu action and confirmation dialog.
JSON file upload now supported in bulk workflows: You can now upload .json files directly to the File Upload pipeline, expanding the range of data formats available for AI workflows.
Bulk processing jobs no longer get stuck on file errors: Fixed an issue where a failed file processing step could leave a bulk job permanently stalled. Workflows now continue processing reliably without needing manual intervention.
Agent input and output schemas now formatted correctly: Resolved an issue where agent input and output schemas were being wrapped with an extra layer, which could cause problems with agent configuration. Schemas are now structured correctly.
Per-run agent instructions now reliably applied: Fixed an issue where additional instructions passed at run time were silently dropped before reaching the AI agent. Agents will now follow per-run guidance as expected.
More reliable authentication sessions: Fixed an issue where refreshing a user token could send an empty client ID, triggering an authentication error and interrupting the session. Sessions should now stay active without unexpected sign-out prompts.
MCP scope filtering now available (feature-flagged): MCP servers can now be configured to filter on a specific subset of scopes via metadata, giving teams more precise control over what data is accessed and retrieved within their workflows.
Improved error handling for MCP tool calls (feature-flagged): Errors from tool calls within MCP workflows are now handled in line with the MCP protocol, resulting in more reliable and predictable agent behavior when a tool encounters a failure.