AI traces vs our ProcessBlock
The gaps from Dimitar's clip are closed and live on all three surfaces
Status
Live on Desktop, ai.allternit.com and m.allternit.com
allternit-ai #686 and platform #1732 are merged, and ai.allternit.com deployed on merge.
The m.allternit.com PWA is deployed (Pages 5ecb4574). The new styles, the bot page and the web proxy all answer.
Desktop needed no new publish. Runtime r20261020.2242 was built from a commit after the merge, and it ships the same process CSS file. Desktop picks it up at its next update check and restart.
What changed
| Gap | Before | Now | Status |
|---|---|---|---|
| App logo on each step | "Used Slack · Search messages" | "Searched [Slack logo] Slack for messages since yesterday" | |
| Typed preview rows | raw input and output | first 3 found items, with sender, one line, time and "+N more" | |
| Older steps close, newest step opens | list closed while working | the list opens while app results arrive, and only the newest step shows its finds | |
| App icon row on the summary | none | "Ran 2 searches, read 3 sources" followed by the app logos | |
| Inline source chips in the answer | plain links | links to Slack, GitHub, Linear, Gmail, Notion and Drive show as chips with the app logo | |
| Full result table during a turn | shown twice, next to the preview | shown once, when the steps finish |
Checks
- Typecheck: 0 errors.
- ai-elements tests: 85 of 85 pass.
- Production and PWA builds pass.
- I checked the live CSS on ai.allternit.com, m.allternit.com and in the Desktop runtime manifest.
- Docs:
guides/agent-steps, the release note and the dependency-map entry are live.