Use cases

Use cases are an index into the calls, not a separate product. Each row names a GTM job, the calls that serve it, and the shape of the decision your code gets back.

Decision shapes. Rewrite returns a better version of a draft, or the draft itself with edits to make. Query returns rows or quotes for you to read. Grade returns a verdict with reasons. Gate returns a send/hold bit. Orchestrate hands a multi-step job to a server-side agent.

With nothing connected

These need only a draft. They are the first results a new workspace can get.

Job	Calls	Decision shape	Start with
Sharpen one draft before it goes out, email or LinkedIn	The Optimizer (messages.optimize)	Rewrite	Optimize a cold email
Before and after on a whole campaign: every step of a sequence, or next week's AI SDR queue	The Optimizer, one call per draft, from a CSV	Rewrite	Optimize a campaign
A style pass in front of an AI SDR or any code that sends	The Optimizer between the writer and the send queue, with a fallback	Rewrite	Optimize before you send
Keep the sender's voice and house rules through every rewrite	The Optimizer with context (rules, voice_examples, facts)	Rewrite	Keep your voice, rules and facts

Once your conversations are synced

Job	Calls	Decision shape	Start with
Catch claims no customer made before a draft goes out	The Optimizer with evidence: "workspace", which lists them in unsupported_claims	Rewrite	Check claims against your customers
Grounded outbound: write a first touch in your buyers' own words	Search (semantic, hydrate: true) for the pain, then Eval (prompt-and-message-eval) on the draft	Query, then Grade	Onboarding skill, use case A
Positioning check: test homepage or pitch copy against what customers say	Search (semantic) for why customers chose you, then Eval with mode: "advisory" and artifact_type: "landing_copy"	Query, then Grade	Onboarding skill, use case B
Agent guardrail: grade what an AI SDR wrote before it sends	Eval with mode: "gate", then the /gate read	Gate	Gate before send
Sequence review: audit every step of a live sequence	One Eval per step, then read the dimensions across steps	Grade	Grade a cold email
Prompt tuning: improve the prompt your team or agent writes from	Eval (default rewrite mode), then an A/B with evidence_from_run	Grade	Pinned-evidence A/B
Objection mining: what did closed-won accounts push back on	Search (filter on deals), then Search (semantic, scoped to those companies)	Query	Cohort, then voice
Funnel and pipeline questions: counts, win rates, cycle time by segment	Search, filter lane with group_by and metrics	Query	Search
Exact-mention checks: did anyone name a competitor or a feature	Search with the term in double quotes (lexical lane)	Query	Search modes
ICP and win/loss research that needs synthesis and outside market signal	A Chat at depth: "deep"	Orchestrate	Chat
Weekly competitive or pipeline sweep	A Routine that fires a Chat on a cron	Orchestrate	Routines
Call prep before a meeting (not available today: the calendar source is paused)	A Subscription that fires a Chat ahead of an event	Orchestrate	Subscriptions
Writing-quality trend across a team	Many Evals, then the KPI read	Grade, aggregated	Eval as a trend

Two limits apply to every row that reads your corpus, which is every row on this page except the Optimizer's default style pass:

• It needs synced conversations. A workspace with none returns empty searches and not_applicable grades, and an Optimizer call with evidence: "workspace" falls back to a style pass. Check GET /search/overview first.
• CRM presence is not a commercial relationship. CRMs create company records from inbound mail, so your corpus holds vendors, recruiters and press alongside buyers. For ICP, voice-of-customer and win/loss questions, scope to accounts with at least one call or deal, and say that you did.

The call-prep row depends on the Google Calendar source, which is the only Subscription source and is not connectable today. See Subscriptions.