JCIL Cloud Docs

Every JCIL Cloud integration uses two keys:

• Private secret — sk_jcil_.... Shown once at creation. Use for server-to-server API calls via Authorization: Bearer or X-API-Key. Never expose this in client code.
• Public ID — pk_jcil_.... Safe to put in HTML. Used for the widget iframe URL and the embed.js data attribute. Protected by the domain allow-list you configure in the console.

For direct API calls (OpenAI-compatible endpoint):

curl https://jcil.ai/api/v1/chat/completions \
  -H "Authorization: Bearer sk_jcil_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jcil-deep",
    "messages": [{"role": "user", "content": "What does Scripture say about forgiveness?"}]
  }'

Or with the OpenAI SDK:

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: "sk_jcil_...",
  baseURL: "https://jcil.ai/api/v1",
});
const res = await client.chat.completions.create({
  model: "jcil-deep",
  messages: [{ role: "user", content: "Hello" }],
});

In the console you pick which models a key can use and which is the default. Per-request override via the model field — but only allowed models work.

The simplest integration. Drop anywhere HTML works — WordPress, Squarespace, Wix, Webflow, raw HTML.

<iframe
  src="https://jcil.ai/embed/pk_jcil_your_key"
  width="400"
  height="600"
  frameborder="0"
  style="border:0;border-radius:12px;box-shadow:0 8px 40px rgba(0,0,0,0.2);"
></iframe>

For a responsive fill-the-container embed:

<div style="position:relative;padding-top:75%;max-width:480px;">
  <iframe
    src="https://jcil.ai/embed/pk_jcil_your_key"
    style="position:absolute;inset:0;width:100%;height:100%;border:0;border-radius:12px;"
  ></iframe>
</div>

For a chat button in the bottom-right corner that opens in a modal — the pattern used by Intercom, Drift, and Zendesk — drop this script anywhere before the closing </body>:

<script
  src="https://jcil.ai/embed.js"
  data-key="pk_jcil_your_key"
  data-position="bottom-right"
  defer
></script>

Options (all data-* attributes):

• data-key — your pk_jcil_... public ID (required)
• data-position — bottom-right (default), bottom-left
• data-label — bubble label, defaults to "Chat"
• data-accent — override the button color (defaults to widget's accent)

The script creates a single floating button. Click opens a modal with the full widget. ESC or click-outside closes it. Fully keyboard-accessible.

The custom_prompt field (2,000 chars max) shapes your assistant's tone, focus, and persona. It's wrapped insideJCIL's base biblical identity block — so you can customize identity, but never override Scripture-aligned values.

Example for a church:

You are St. Paul Baptist Church's AI assistant.

Tone: warm, pastoral, hospitable.
Format: answers under 120 words when possible, offer next steps.
Sign-off: "— Pastor Mike's AI helper"

If someone asks about:
  • Service times: link to /services
  • Visiting: offer to schedule with our hospitality team
  • Prayer: acknowledge warmly and offer the prayer request form
  • Doctrine: keep answers simple, point to Pastor Mike for depth

Never discuss politics unless scripturally relevant.

Example for a Christian university:

You are the admissions assistant for Providence Christian University.

Tone: professional, encouraging, student-focused.
Audience: prospective students, parents, current students.
Format: direct answers with specific links to admissions, financial aid,
or the registrar when relevant.

Key pages you should reference:
  • /admissions for application details
  • /financial-aid for FAFSA + scholarships
  • /academics for degree programs
  • /student-life for housing, chapel, activities

Do not speculate on policies — if unsure, direct to
admissions@provchristian.edu.

Values you cannot override with your custom prompt:

• Faithful to Scripture and the Gospel of Jesus Christ
• Pro-life in every phrasing
• Refusal to produce sexually explicit or blasphemous content
• Refusal to affirm teachings that directly contradict the Bible
• Points users in crisis to real human support (counseling, pastoral care)

Provide up to 10 URLs. We fetch each page, strip nav/footer/scripts, cache the main content for 1 hour, and inject it into every chat request. Budget: 60KB of text total across all sources (~15K tokens).

Great candidates:

• About / Who We Are
• Statement of Faith / Beliefs / Doctrine
• Service times / Events calendar
• Staff directory / Contact page
• Admissions / Program pages (universities)
• FAQ pages
• Ministry descriptions
• Give / Donate page

When it's time for real RAG:if your knowledge base grows past 100 pages, reach out — we're rolling out pgvector-backed embeddings for Enterprise customers.

Base URL: https://jcil.ai/api/v1

POST /chat/completions

OpenAI chat completions format. Supports streaming via stream: true.

{
  "model": "jcil-deep",         // or "jcil-standard", "jcil-fast"
  "messages": [
    { "role": "system", "content": "..." },  // optional — merged with JCIL base
    { "role": "user", "content": "What time is Sunday service?" }
  ],
  "stream": false                    // set true for SSE streaming
}

Response (non-streaming)

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1776400000,
  "model": "jcil-deep",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "..." },
    "finish_reason": "stop"
  }]
}

Response (streaming)

Server-Sent Events (SSE). Each chunk is a data: {...} line with an incremental delta. Terminated by data: [DONE].

GET /chat/completions

Returns the available model IDs. Lets OpenAI SDK clients probe compatibility.

• Per widget visitor — 10 messages/min, 100/hour per IP. Prevents abuse from burning your credits.
• Per API key (Developer) — 60 requests/min.
• Per API key (Business) — 300 requests/min. Email support@jcil.ai to upgrade.
• Per organization (Enterprise) — custom. 1500+/min available.

Rate limit responses return HTTP 429 with a Retry-Afterheader. If Upstash Redis is unavailable, we fail open — you won't be blocked by infra hiccups.