Your bot. Our policy engine.
Wiki Connect is an HTTP API for policy-constrained, citation-backed answers. Keep the chatbot you already have. Stop it from inventing a refund.
An API key on the application. A POST for the question.
Wiki Connect does not replace your support chatbot. It governs the knowledge that chatbot may use, then returns a cited answer or a refusal.
Most teams already have a bot. The failure is not the widget. It is the unreviewed folder behind it. Wiki Connect sits on an AI Application you configure with verified sources, public-safe or internal collections, and a policy.
Your bot calls POST /api/v1/answers with a Bearer key. The response includes the answer, citations, status, refusal reason, and decision. Native Intercom or Zendesk apps are not included.
Six live steps from a verified source to a cited answer.
- Step 1
Create an AI Application
Model the public chatbot or internal copilot as its own application with a policy over collections.
- Step 2
Verify the sources it may cite
Import a URL or write the document, then mark it verified. Unreviewed pages stay out of answers.
- Step 3
Keep public-safe and internal collections split
Customer-facing bots stay on public-safe knowledge. Employee copilots stay on internal collections.
- Step 4
Test in the Answer Playground
Ask the refund, billing, and edge-case questions first. Confirm citations, or confirm a refusal.
- Step 5
Create a Wiki Connect key
Open Wiki Connect on the application, name the key, and copy the secret. Wiki stores only a hash.
- Step 6
POST /api/v1/answers
Send Authorization: Bearer wk_... and a JSON body with question. Read answer, citations, status, and refusalReason.
What Wiki Connect actually does.
These are shipped capabilities. Source change detection, stale-source alerts, and native helpdesk apps are planned and not claimed here.
Policy-constrained retrieval
The application policy is the allow-list. The model does not get a master key to the workspace.
Citation-backed answers
When an approved source matches, the response includes the citation your team can inspect.
Honest refusals
When nothing approved matches, the API can refuse instead of guessing.
Same engine as the Playground
The question you tested is the question your bot can serve. No second prompt stack.
Public-safe versus internal
Collection visibility stays enforced at answer time, including through the HTTP API.
Your bot, your UI
Wiki Connect is an HTTP API. There is no native helpdesk app to install.
Questions before you generate a key.
- What is Wiki Connect?
- An HTTP API on each AI Application. Your bot sends a question with a Bearer key and receives a citation-backed answer, a refusal, or a decision from the same policy engine as the Answer Playground.
- Does Wiki Connect replace my chatbot?
- No. Keep your existing bot. Wiki Connect is the knowledge and policy layer behind it. Native Intercom or Zendesk apps are not included.
- What happens when there is no approved source?
- The API can refuse. A missing policy is a knowledge gap, not a reason to invent a refund or a shipping exception.
- Can a public chatbot retrieve internal collections?
- No. Public-facing applications stay scoped to public-safe collections. Internal notes stay on internal collections and the public bot cannot use them.
- How do I try it on the $1 trial?
- Create a workspace, verify a public policy URL, attach an AI Application policy, test the question in the Answer Playground, then create a Wiki Connect key and call POST /api/v1/answers.
Docs, use cases, and the refund walkthrough.
Your AI is already talking.
Make sure it knows what it is talking about.