Source-backed drafts
Responses remain proposals and identify the approved policy boundary used.
All rolesSandbox
Customer Relations
Handles bounded service conversations using approved knowledge, with human takeover and opt-out built in. This release creates response drafts only: no message is sent to an external customer.
A
Synthetic sandbox workerAshaCustomer Relations associate
Draft-only- Primary outcome
- Source-backed service reply drafts
- Authority
- Classify and draft in approved scope
- External action
- Disabled in this release
- Human control
- Takeover, escalation and opt-out enforced
Sandbox charter
A bounded conversation assistant, not an autonomous service desk.
Asha works only from an approved FAQ and policy library. Participant references are hashed, suppressions are enforced before drafting, and a human can take over at any time.
Included
- Intent triage and source-backed reply drafts
- Approved service responses across supported channels
- Human takeover and conversation escalation
- Inbound intent and urgency classification
- Draft provenance and policy-limit disclosure
- Hashed opt-out and do-not-contact suppression
Not included
- Sending email, chat, SMS or messaging-app replies
- Changing accounts, refunds, credits or contractual terms
- Handling threats, regulated advice or identity verification
- Continuing after human takeover or participant opt-out
- Using knowledge outside the approved service library
Conversation controls
Every path has a visible stop condition.
Immediate takeover
A human takeover closes the drafting path server-side, not merely in the interface.
Exception escalation
Sensitive, unclear or prohibited requests are routed to a named human owner.
Suppression first
Opt-out is checked before work begins and stores only a one-way participant hash.
Sandbox evidence
Open the sample workspace Inspect the control path from the employer side.
Create an inbound synthetic thread, generate a draft, take it over and verify that further AI drafting is blocked.