Wonderchat

Custom tools call external HTTP endpoints with API key or bearer authentication. Builders can use GET, POST, PUT or DELETE, set typed parameters, mark required values and supply defaults. A tool can be invoked from a conversation workflow, which uses the returned data in its next response. The setup includes a test request and response inspection before saving, allowing customers to connect both read operations and changes in their business systems.

Sourcedocs.wonderchat.io/setup-guides/using-chatbot-tools/set-up-custom-chatbot-toolsread 2026-10-20

A workflow starts from a condition the AI detects in a customer's message and continues through an ordered set of actions. Builders add fields to collect, API calls and response steps that consume the collected values or tool result. The product also supports branching on the visitor's answers for qualification, routing and escalation. Customers edit this conversation process and can launch a preview from the workflow's Play control before putting it into use.

Sourcedocs.wonderchat.io/setup-guides/adding-chatbot-workflows/set-up-chatbot-workflowsread 2026-10-20

Knowledge sources include website pages, PDF and Word documents, audio or video, manual text, Google Drive and SharePoint. Teams can also import support ticket knowledge. Website refreshes run manually or on a plan dependent daily, weekly or monthly schedule. The answer depth setting changes how much source material retrieval considers. Source citations and stored corrections give staff ways to verify responses and improve the content used for subsequent questions.

Sourcedocs.wonderchat.io/faq/chatbot-technical-faqread 2026-10-20

Staff can receive an escalated conversation with its history, continue it through live chat or handle it as an email or help desk ticket. Instructions limit the topics and information an agent shares, and sensitive topics can be routed into approval workflows. The routine operating surface centers on handover, response rules and corrections. External API tools do not have a general per-call approval queue with a configurable reviewer and release action.

Sourcedocs.wonderchat.io/faq/chatbot-technical-faqread 2026-10-20

Account owners, managers and users hold different permission levels, including roles that limit who sees personal information in conversations. Enterprise plans add single sign on, and users can also sign in through supported identity providers or a magic link. Wonderchat holds SOC 2 Type II and offers a HIPAA business associate agreement, with SOC 2 and HIPAA listed as Enterprise plan features. Databases, archives and logs are encrypted at rest, connections use HTTPS, and model provider arrangements keep customer chats out of model training.

Sourcetrust.wonderchat.io/securityread 2026-10-20

Operators can review full conversation transcripts, search the exchanges and export selected chat records, with access controls governing who reads them. Citations identify the source material behind each answer, feedback and corrections mark the replies staff fixed, and analytics highlight repeated questions and knowledge gaps, with advanced analytics on Scale. The record covers what was said and which sources were cited; individual tool requests and responses made in production do not appear as steps in it.

Sourcewonderchat.io/feature/chat-logsread 2026-10-20

A returning visitor is recognized and the agent picks up from the earlier conversation, including an unresolved question and where the discussion stopped. The customer sets how long chat history is kept and when a conversation resets, and Enterprise adds custom retention policies and role based access to conversation history. What the agent recalls is that visitor's stored chat history, managed by exporting or deleting the chat logs themselves, and CRM connections supply the customer record. Factual answers still come from the configured knowledge sources.

Sourcewonderchat.io/feature/chat-historyread 2026-10-20

The service runs in Wonderchat's managed cloud, and its solution pages give EU hosting in Frankfurt as the location, with no region selector in the account. Enterprise customers can negotiate retention policies and connect private databases as knowledge sources, but the agent runtime still runs on Wonderchat's infrastructure, and the standard service offers no package to install in a customer's own cloud or data center. Embedding the widget or connecting WhatsApp, SMS, Slack or Teams changes where customers reach the agent, not where it runs.

Sourcewonderchat.io/solutions/conversion-agentread 2026-10-20

Three named agents are sold as ready starting points for different jobs: a Support Agent that answers from help content and routes unresolved requests to people, a Conversion Agent that answers buying questions, qualifies visitors and books meetings from the customer's own content, and an Internal Agent for employee knowledge questions. Every plan, the free one included, carries the Conversion Agent alongside support. Customers add their own sources, instructions and tools to each, and both customer facing agents share the same knowledge and workflow configuration.

Sourcewonderchat.ioread 2026-10-20

Incoming customer messages can trigger a workflow when the AI recognizes a configured condition. The agent then collects information or invokes its assigned actions without a staff member starting each run. Customers reach the service through website widgets and connected messaging channels, with application access through the chat API. Dynamic greetings and routing use conversation context, while outbound webhook events notify connected systems when activity occurs.

Sourcedocs.wonderchat.io/setup-guides/adding-chatbot-workflows/set-up-chatbot-workflowsread 2026-10-20

Customers select the model that powers each chatbot, with choices from OpenAI, Anthropic, Google, xAI and other providers, trading response speed, cost and reasoning capacity while the knowledge configuration stays the same. Answer depth has named presets with a visible credit cost, so retrieval breadth is set alongside the model, and message cost varies with both. A context window setting lengthens memory within a conversation for dense, multipart questions or shortens it for speed and lower cost.

Sourcedocs.wonderchat.io/faq/chatbot-technical-faqread 2026-10-20

The REST API accepts authenticated requests using an account API key and a chatbot identifier. Endpoints send messages to an agent, add or remove indexed pages, export conversations, retrieve messages and manage chat tags and feedback, so a developer can run the hosted agent inside an application, keep its knowledge in sync with a content system and pull transcripts into a data warehouse. Webhooks sign outgoing event payloads with a shared secret so the receiving system can verify their origin.

Sourcedocs.wonderchat.io/api-reference/api-referenceread 2026-10-20

The evaluation page lets a customer select an existing chatbot, enter a test question and run it against one or more models at once, with the results shown together so accuracy, grounding, tone and safety behavior can be compared on that chatbot's own knowledge. Workflow previews start from a workflow's Play control, and custom tools have a test request with response inspection before saving. Comparisons are read by a person one question at a time; no saved test set or automatic score sits behind them.

Sourcedocs.wonderchat.io/setup-guides/adding-chatbot-workflows-1/how-to-use-the-single-prompt-evaluation-featureread 2026-10-20

Smart Navigation answers a visitor's question from the relevant product, policy or FAQ page and supplies a link to that source. Private crawling can ingest content behind a login for the knowledge collection. Neither feature operates a customer browser on behalf of the agent. Business actions use defined API tools, and visitors follow links or interact with the chat widget themselves; the runtime has no general screen clicking or desktop control engine.

Sourcewonderchat.io/feature/smart-navigationread 2026-10-20