Launch evidence
Proof before expansion
ZChat packages launch review around manual evidence: regression scenarios, handoff transcripts, channel continuity checks, rollback references, and owner approval before automation widens.
Manual regression packet
Review stale knowledge, sensitive actions, identity continuity, failed tool runs, and replay candidates before changing prompts, models, or channels.
- Scenario matrix
- Replay candidates
- Reviewer notes
Handoff readiness
Show that visitor transcript, AI reply, handoff summary, and first human response are visible before citing fallback readiness.
- Transcript
- Handoff summary
- Human response
Expansion boundaries
Separate shipped channel evidence (website, email, SMS, messaging apps, voice) from unshipped ChatGPT or Gemini assistant surfaces and native mobile apps.
- Current surface
- Planned surface
- Missing proof
Expansion waits for evidence, rollback, and owner approval.
| Capability | Status | Detail |
|---|---|---|
| AI distribution readiness | API only | Launch packets collect prompt, model, knowledge, tool, channel, routing, and audit-event references. They come from the dashboard API (/api/app/ai-distribution/...); no dashboard screen shows them. |
| Manual regression run packet | API only | Scenario count, replay transcript ids, failed scenario count, reviewer notes, and rollback reference, from the dashboard API; no dashboard screen shows it. Answer tests and scenarios themselves run on the Test AI answers page. |
| Human handoff readiness | Available | Review handoff transcripts, stale queue rows, and first-human-response evidence before widening AI fallback. |
| Channel continuity review | Configured | Email, messaging apps and the AI phone agent run on accounts you connect. Send a test conversation through each one and check it reaches the inbox with its history before customers rely on it. |
| Automated eval harness | Available | Save real customer questions and re-run them through the live grounded answer path after knowledge or prompt changes; pass/fail results persist per question. Not automated prompt rollback, provider-certified QA, or guaranteed outcomes. |
| External agent coverage | Not available | Not available: ChatGPT or Gemini app listings, native mobile apps, or an app marketplace. Integrations lists the channels that work today. |
This is the evidence layer broader AI-agent suites often blur: ZChat should expand only after review artifacts and rollback ownership are visible.