Sharper Chat Audits, Smoother Chat, and Steadier Runs ​

July 7, 2026 · InfraScout Team

This release is a roundup of smaller improvements across three areas: the admin Chat Audit reports what happened more accurately, the chat view itself picks up a round of polish, and a handful of behind-the-scenes fixes make workflow runs, agent updates, and web research quietly more reliable. None of them change how you work — the same views and tools just behave a little more precisely.

A more accurate Chat Audit ​

The admin Chat Audit table gains a Model column, so you can see which model answered a conversation without opening the full transcript. That is most useful once you are running some connections on Anthropic and others on OpenAI: the column tells you at a glance which model handled a given conversation, and where a single conversation involved more than one, it reflects that too.

The Cost column added last release gets more trustworthy alongside it. It now reflects accurate spend for Agentic Workflow runs, not just interactive chats, so an automated assessment's cost reads the same way an interactive one does. And conversations old enough to predate some of today's cost tracking now render cleanly instead of showing a blank or broken row, so the table stays readable all the way back through your history.

A round of chat polish ​

The chat experience picks up several smaller refinements. Each is minor on its own, but together they smooth over rough edges you may have run into recently.

  • Mid-loop steering — sending a follow-up while the AI is still working — now behaves identically whether the connection runs on Anthropic or OpenAI models, so steering a running assessment feels the same on either provider.
  • The conversation view now fills the full height of the window, and the Chat item in the sidebar stays highlighted while you are inside a conversation, so you always know where you are.
  • The playbook and provider pickers you use when starting a new chat are now richer, easier-to-scan dropdowns.
  • Long tool-call arguments stay inside the chat panel instead of overflowing it, so a call with a large payload no longer breaks the layout.

Steadier behind the scenes ​

A few reliability fixes round things out, mostly in places you do not look at directly but notice when they go wrong.

  • Agentic Workflow runs start faster and report failures more clearly, so a run that goes wrong tells you what happened instead of leaving you guessing.
  • Agent auto-update is more dependable: an update that has finished no longer lingers as still in progress, each agent installs exactly the build that was requested, and an update that stalls partway now heals itself.
  • Command-type badges in the Executions list stay inside their column instead of overflowing it.
  • Domain allow and block lists for AI web research now match bare hostnames reliably, so an entry that is slightly malformed no longer silently fails to take effect.

Questions or feedback? Reach us at info@infrascout.cloud.