Fuel iX Copilots — Week 39

This week brought a mix of new capabilities and reliability fixes — from smarter conversation summaries and new file support to more accurate usage tracking and smoother session handling across chats and agents.

Automatic conversation summaries: Long chat conversations now generate summaries automatically, so it's easier to catch up on context and key points without scrolling back through the full history.
XML file upload support: You can now upload and process XML files directly in Fuel iX, giving you more flexibility in the data sources you bring into your AI agents and workflows.
Cache usage visibility in analytics: Usage analytics now show prompt cache hit rate as a percentage of input tokens in the model breakdown, helping teams understand and optimize caching efficiency.
More accurate token usage counts: Fixed an issue where the conversation usage panel could underreport token usage by up to 50% due to bad data from providers — token counts shown now are validated and accurate.
Corrected default model access for new organizations: New organizations now start with the recommended default set of AI models instead of being given access to every model, for a more curated starting experience.
Copilot thumbnail updates fixed: Copilot creators can now update thumbnail images on their existing custom copilots without needing super admin permissions.
Smoother recovery from expired sessions: When a session expires, the system now gives clearer feedback and triggers a proper refresh flow, instead of showing a generic error.
Chat no longer interrupted when editing copilot instructions: You can now update a copilot's instructions mid-conversation without the chat failing or shutting down.
Clearer handling when prompts are blocked: If a prompt is blocked by safety guardrails, the agent run now properly shows as failed with a clear message instead of getting stuck in a "queued" state indefinitely.
Better visibility into agent workflow retries: Agent execution monitoring now reports retry attempts more accurately, making it easier to troubleshoot workflow issues.