Chat Personalization

The first-run onboarding inside chat. A user with no row is asked "Who am I teaming up with?" and fills one sentence, I'm a [role] at [company]. Answering queues background research that reads the company and has a model author use-case cards, which then replace the stock cards in the empty state. Lives with chat, so it is Cloud and Enterprise only and never reachable when chat is not.

UNSET the synthesized status meaning no row exists, so this person has never been asked. Never persisted. DISMISSED_LEGACY written by the shipping backfill to every user who existed before the feature. Terminal, and deliberately distinct from SKIPPED. SKIPPED the user saw the question and declined, or research was not allowed to run. Company row the userId IS NULL row holding the company answer and its research; one per platform. User row a per-person row; a teammate's own role-targeted result wins over the company row once READY.

How it works

• GET/POST /v1/agents/personalization, mounted on the chat surface behind chatVisibilityGuard.
• POST writes the answer, runs guardsAllowResearch (chat AI provider present, credits not blocked, under 5 runs per platform per day) and queues EXECUTE_PERSONALIZATION_RESEARCH. It returns 202 with the current view; progress arrives over CHAT_PERSONALIZATION_PROGRESS on the userId room, and the client also polls while the status is in flight.
• The worker claims the row by flipping PENDING to RESEARCHING, gathers a homepage read plus web searches, then synthesizes a profile and 12 to 20 cards in parallel on the fast model. A crashed run is recovered on read: an in-flight row whose heartbeat stopped two minutes ago is reset and re-enqueued, bounded by the daily cap.
• The company blank prefills from platform.name, not from the user's email, via chatPersonalizationUtils.companyFromPlatformName. That covers invited teammates whose personal address does not match the company. It returns null for a name the signup generator produced from a person, so the blank stays empty rather than offering "Ahmad's Platform" as a company.

Gotchas

• The prefill only works once the platform is named after the company. It reads platform.name, so it depends on work-email platform naming; before that shipped every platform was "<FirstName>'s Platform" and the predicate correctly rejected all of them. A person-named platform yields an empty blank, not a wrong one.
• Apollo and Clearbit are Cloud-only and both optional. Apollo guesses the role from AppSystemProp.APOLLO_API_KEY and Clearbit powers company autocomplete from the browser. Neither runs off Cloud, and neither is required: the company blank fills locally on every edition. Apollo deliberately does not read an AI-tool config, because those are per-platform and a fresh signup has none.
• Research needs a chat AI provider or it degrades to SKIPPED silently. guardsAllowResearch returns false with no provider and the user simply keeps the stock cards. See the AI Providers page for why a local Cloud platform has no provider until you flip aiProvidersEnabled.
• A backfilled user never sees the card, so the personalization chip is their only way in. The card is gated on UNSET, and the migration made every pre-existing user DISMISSED_LEGACY. Removing the chip would lock the entire existing base out of the feature permanently.

Key files

• packages/server/api/src/app/ee/agent/personalization/ — entity, service, controller
• packages/server/worker/src/lib/execute/jobs/ee/agent/execute-personalization-research.ts — the research run
• packages/core/shared/src/lib/ee/agent/chat-personalization.ts — statuses, view, chatPersonalizationUtils
• packages/web/src/features/chat/lib/ — use-personalization.ts, onboarding-prefill.ts, personalization-api.ts
• packages/web/src/features/chat/use-cases/ — the card set and its code-drawn art
• packages/web/src/app/routes/chat-with-ai/components/ — onboarding-question-card.tsx, onboarding-welcome.tsx, personalization-chip.tsx