Better Threads, Broader Reach

This week focused on the parts of Humm that shape day-to-day usage: getting more of your data into the product, keeping thread navigation stable, and making generated workflows feel less fragile. You should see cleaner thread behavior, a wider set of sync options in the integrations catalog, and fewer moments where Humm loses its place.

16 New Data Sync Connectors

Humm now supports 16 more data sync connectors:

  • Basecamp
  • Customer.io
  • Delighted
  • Freshdesk
  • GitLab
  • Help Scout
  • Meta Ads
  • Monday
  • Paddle
  • Productboard
  • Recurly
  • Statsig
  • SurveyMonkey
  • Trello
  • Typeform
  • Zoho CRM

Once connected, these sources flow through the sync pipeline into queryable data sources for analysis and reporting.

Threads That Keep Their Place

Deep-linked threads now behave much more consistently when you open them with an orgId, switch organizations, or refresh the sidebar. We also fixed thread ordering and duplicate-thread edge cases, so the list you see is closer to the thread state you actually expect.

Commands and Clarifications That Finish Cleanly

Creating a command from a thread is now more reliable across navigation, with clearer generating and failed states while the command is being built. You can move around the app without losing the generation state, and the Commands screen now updates when the command is ready.

We also fixed ask-human clarification flows so custom Other answers can continue normally instead of stalling the interaction. Small change, very noticeable when you're in the middle of a thread and just trying to answer the question in your own words.

Cleaner Agent and Tool Output

Tool results now arrive in a more structured format, which gives the UI a better shot at showing failures, auth issues, empty results, and successful runs correctly. On the user side, that means clearer tool-call states inside threads and less ambiguity about whether Humm actually got what it needed.

We also tightened GPT-5.4 prompt and routing behavior across the analyst stack. The change is under the hood, but the intended outcome is visible: better tool selection, better recovery from tool errors, and less wandering in complex analyses.

Improvements

  • Data tables now default to 10 rows per page instead of 25, which cuts down on visual clutter in larger result sets.
  • Sidebar and profile UI got a small cleanup pass across link colors, chips, avatars, and organization-name overflow.
  • Intercom conversation search was tightened up to be more reliable for larger queries.

Bug Fixes

  • Fixed deep-link thread handling when switching organizations, including sidebar refresh and selected-thread recovery.
  • Fixed duplicate thread creation after proxy timeout retries.
  • Fixed command generation state getting lost across navigation.
  • Fixed clarification cards so custom Other responses can continue instead of blocking submission.
  • Fixed Intercom query timeouts caused by an overly aggressive proxy path.