Capabilities
Current scope and maturity
ZChat publishes what ships today, channel by channel and feature by feature, so buyers can separate shipped behavior from planned expansion and unclaimed suite parity.
Available now
An AI agent grounded in your knowledge, 13 built-in actions, flows, a shared inbox for 12 channels, same-thread human handoff, and your choice of AI provider or key.
- AI agent and actions
- 12-channel inbox
- Provider choice
Connected with your accounts
Messaging channels and your own API tools run on credentials you supply. Give each one a named owner and test it before customers rely on it.
- Owner
- Payloads
- Rollback
Not claimed yet
Large marketplace breadth, formal compliance badges, and deep CRM-native workflows remain outside ZChat's public parity claims.
- No fake marketplace
- No badge without proof
- No CRM-suite claim
Capability status
| Capability | Status | Detail |
|---|---|---|
| Website chat widget | Available | Install script, capture first intent, keep human handoff visible, and review conversations in the inbox. |
| Human-owned inbox | Available | Operators own follow-up, use saved views, and keep resolution context attached to the conversation. |
| Approved knowledge drafting | Available | AI answers should be grounded in approved sources and escalate uncertainty or sensitive requests. |
| Customer-managed AI provider mode | Available | Customers can separate platform spend from model usage when finance or security review needs provider clarity. |
| HTTPS tools and sensitive actions | Configured | Point the assistant at endpoints you define; every call is logged with arguments, status, response, and latency. ZChat does not hold calls for approval, so put that check inside the endpoint before it changes an account, refund, or entitlement. |
| AI tool calling with a per-call audit trail | Available | Define HTTPS tools per workspace and the assistant calls them mid-conversation, serially and with a round cap. Every call records arguments, response status, body, error, and elapsed time. Off until you enable it. |
| Sentiment auto-escalation | Available | Each visitor message is scored and rolled into a per-conversation average. When it falls below your threshold the conversation moves to a human before the model is called again, so an unhappy customer stops receiving AI replies. |
| Golden-question regression evals | Available | Save real customer questions and re-run them after a knowledge or prompt change to catch regressions before visitors do. Runs on a side-effect-free preview path, so nothing lands in conversations or usage. |
| Signed visitor identity | Available | Your server signs an identity token with the workspace HMAC secret. The assistant then knows it is talking to a verified signed-in user and may use authenticated tools on their behalf. |
| Replies in the visitor's language | Available | Reply language follows the visitor's browser, with a workspace default behind it and an explicit override in front. No per-language configuration to maintain. |
| Conversation summaries and agent drafts | Available | Summarise a long thread before picking it up, and generate a draft reply the agent edits. Drafts are never sent automatically. |
| Auto-tagging and topic tagging | Available | Conversations are tagged automatically, so reporting reflects what customers actually asked about without operators remembering to tag. |
| Agent scorecards and daily digest | Available | Per-agent performance views, plus an optional once-a-day digest email to the workspace notification address. |
| Operator mentions and message reactions | Available | Pull a colleague into a conversation with an @mention, and react to messages in the shared inbox. |
| Built-in AI actions | Available | Thirteen actions the AI can take without you hosting an endpoint: leads, custom fields, hand-off, tickets, Zendesk or Freshdesk escalation, Shopify orders and products, reply and link buttons, booking, Slack, web search and Stripe billing. Each is off until enabled and every call is logged. |
| Flow builder | Available | Step-based conversation flows with buttons, questions, branches, tags, actions and hand-off to the AI or a person. |
| Messaging channels and voice | Available | Email, SMS, WhatsApp (Twilio or Meta Cloud API), Messenger, Instagram, Telegram, Discord (/ask command), LINE, Microsoft Teams, Slack, an AI phone agent on Twilio Voice, and one-off email or SMS campaigns. Each uses your own account credentials, stored encrypted. |
| Multiple sites and agents | Available | Paid workspaces can run up to 10 sites, each with its own agent name, instructions and knowledge scope. |
| Single sign-on | Available | OpenID Connect SSO for operators, matched by email domain, with an option to require it. |
| Native marketplace integrations | Planned | Do not market native OAuth or marketplace parity until scopes, auth boundaries, and rollback behavior ship. |
| SOC 2, GDPR, SLA, and residency badges | Evidence required | Current trust copy can describe controls and decisions; formal badges need signed proof. |
Use this page as the source of truth when comparing ZChat to broader suites: shipped behavior first, roadmap language second, no unsupported parity claims.