The API

Everything HelpPuff does is available over HTTPS, so you can use it as a backend only, without the widget or the dashboard:

• Chat: start a conversation from your own server or app, send the visitor's messages, get the assistant's replies (as JSON or streamed). The same assistant, knowledge base, limits, lead capture, summaries and webhooks as the widget.
• CRM: leads (create, update, export, delete, "forget me"), callback requests, conversations.
• Everything else: the knowledge base, the prompt, settings, webhooks, analytics, the team, API keys and the audit log.

This page explains keys, conventions and the main flows. API reference lists every endpoint with a curl command, the request, an example response and its errors. The same description, as OpenAPI 3.1 (for Postman, Insomnia or a client generator), is at https://<your worker>/api/v1/openapi.json.

Quickstart

1. Make a key. Dashboard → Settings → API keys → New key, or:

   helppuff keys create "My backend" --preset chat

   Copy it: it is shown once.

2. Start a conversation (your Worker's address is in helppuff status, or Settings → API keys):

   export HELPPUFF_URL="https://knowtific-helppuff-acme.you.workers.dev"
   export HELPPUFF_API_KEY="hp_live_…"
   
   curl -X POST "$HELPPUFF_URL/api/v1/conversations" \
     -H "Authorization: Bearer $HELPPUFF_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{"message": "Do you work weekends?", "contact": {"name": "Ada", "email": "ada@example.com"}}'

   {
     "id": "1b0f6a52-3c1e-4c55-9f0e-2d1c7a9e5b10",
     "messages": [{ "id": "m_mfx2k1", "ts": 1760000000000, "role": "agent", "type": "text", "text": "Yes, Saturdays 8am to noon…" }],
     "externalId": null,
     "metadata": null
   }

3. Continue it with the id:

   curl -X POST "$HELPPUFF_URL/api/v1/conversations/1b0f6a52-3c1e-4c55-9f0e-2d1c7a9e5b10/messages" \
     -H "Authorization: Bearer $HELPPUFF_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{"text": "Great, can someone come this Saturday?"}'

The conversation, the lead (Ada) and any callback request appear in the dashboard and go to your webhooks, as if Ada had used the widget.

JavaScript

const base = `${process.env.HELPPUFF_URL}/api/v1`;
const headers = { Authorization: `Bearer ${process.env.HELPPUFF_API_KEY}`, 'Content-Type': 'application/json' };

async function call(method, path, body) {
  const response = await fetch(`${base}${path}`, { method, headers, body: body && JSON.stringify(body) });
  const json = await response.json();
  if (!response.ok) throw new Error(`${json.error.code}: ${json.error.message}`);
  return json;
}

const chat = await call('POST', '/conversations', { message: 'Do you work weekends?', externalId: 'user-123' });
const next = await call('POST', `/conversations/${chat.id}/messages`, { text: 'Can someone come Saturday?' });
console.log(next.messages.map((m) => m.text).join('\n'));

Python

import os, requests

base = os.environ["HELPPUFF_URL"] + "/api/v1"
headers = {"Authorization": f"Bearer {os.environ['HELPPUFF_API_KEY']}"}

chat = requests.post(f"{base}/conversations", json={"message": "Do you work weekends?"}, headers=headers).json()
reply = requests.post(f"{base}/conversations/{chat['id']}/messages", json={"text": "Can someone come Saturday?"}, headers=headers)
reply.raise_for_status()
print([m.get("text") for m in reply.json()["messages"]])

Keys

Make keys in the dashboard (Settings → API keys) or with the CLI (helppuff keys create|list|revoke). Each key:

• works on one site (a deployment made by the CLI has exactly one);
• has scopes: what it may do (below). Presets: Chat only, CRM, Read-only, Full access;
• may expire (30 days, 90 days, a year, or never), may be limited to IP addresses or ranges, and has its own requests a minute (default 120, security.limits.apiRequestsPerKeyPerMinute);
• looks like hp_live_k7m3p9q2r4s8_…. The part up to the second _ is its public id, shown in lists; the rest is the secret.

Send it on every request, in the header and nowhere else:

Authorization: Bearer hp_live_k7m3p9q2r4s8_…

A key in the URL (?api_key=…) is refused: URLs end up in logs.

Scopes

Scope	Lets the key
chat	Talk to the assistant as a visitor: start conversations, send messages, close them
conversations:read / :write	Read conversations and messages / summarise and delete them
leads:read / :write	Read and export leads / create, update and delete them
callbacks:read / :write	Read callback requests / mark them done or dismissed
knowledge:read / :write	Read pages, files, facts and search / crawl, upload, add knowledge, correct facts
prompt:read / :write	Read the prompt and its versions / publish and restore
settings:read / :write	Read / change the settings
webhooks:read / :write	Read / add, change, test and remove webhook endpoints
analytics:read	The overview, usage and the running version
team:read / :write	Read / add and remove dashboard accounts, make sign-in links
keys:read / :write	Read / create and revoke API keys
audit:read	Read the audit log

:write includes :read. A key can never create a key with scopes it does not have itself. team:write and keys:write can give access to others: treat a key that has them like a password to the dashboard.

Keeping keys safe

• Server only. Never put a key in a web page or a mobile app: anyone can read it there. The API sends no CORS headers, so browsers cannot call it. For your own chat UI in a browser, see a custom frontend.
• Shown once, stored hashed. HelpPuff keeps only a keyed hash (HMAC-SHA-256, keyed from your Worker's HELPPUFF_SECRET, which is not in the database). A copy of the database cannot be used to call the API, and a lost key cannot be recovered: revoke it and make another.
• Rotate by making a new key, switching your server to it, then revoking the old one. A revoked key is refused everywhere within 30 seconds.
• Least privilege: give each integration its own key with only the scopes it needs, an expiry, and the IP addresses it calls from when you know them.
• Audit: every change made with a key (and in the dashboard) is in the audit log, with the key's id: GET /audit.
• Rotating HELPPUFF_SECRET invalidates every API key (and signs everyone out). Make new keys afterwards.

The CLI and agents can also use the deployment's own ADMIN_API_KEY (in your project's .env), which has full access: helppuff api GET /leads.

Chat as a backend

POST /conversations starts a conversation as a visitor would, and POST /conversations/{id}/messages continues it. What is the same as the widget: the assistant, its instructions and knowledge, rich messages, lead capture (from contact and from what the visitor types), callback requests, summaries, webhooks, the per-conversation and daily caps. What differs: there is no Origin check or Turnstile (the key is the proof), and the key's own rate replaces the per-visitor limits.

• Who the visitor is: contact (name, email, phone, any other fields) becomes a lead, or joins the existing lead with that email. externalId is your own id for the conversation or visitor (GET /conversations?externalId=… finds it again); metadata is stored and returned, never shown to the assistant.
• Where they are: context (pageUrl, pageTitle, locale, timezone, utm) is what the widget would send.
• Rich replies: messages are the widget's protocol objects (text, options, card, carousel, links, form, notice; see Protocol). Render what you support; show text otherwise.
• Answering a rich message: send {"action": {"id": "<message or action id>", "value": "…"}}. A form's answers go in value as a JSON object string, and are accepted only for a form this conversation was shown.
• Streaming: send Accept: text/event-stream. You get delta events ({"text": "…"}) as the reply is written, then one done event with the same body the JSON answer has (or an error event). Backends that cannot stream answer with plain JSON.
• Ending: POST /conversations/{id}/end (sends conversation.ended). Conversations also complete on their own 5 minutes after the last message (conversation.completed, with the summary).
• Async work: use Webhooks (lead.captured, callback.requested, conversation.completed …) instead of polling.

A custom frontend

To build your own chat UI in a browser, keep the key on your server and let your server make the calls: your frontend talks to your server, your server to HelpPuff. (A short-lived browser token for your own frontend is planned.)

Conventions

• Base URL: https://<your worker>/api/v1. JSON in, JSON out (UTF-8). Times are milliseconds since the epoch (UTC).
• Site: a key always works on its own site; site in a body or query is optional, and another site's value is refused (403).
• Lists: GET /conversations and GET /audit page with a cursor: pass the answer's next as before. Other lists return up to 500 items with counts.
• Errors: always {"error": {"code", "message", "retryAfter?"}} with the HTTP status. message is safe to show a person. Codes: bad_request 400, unauthorized 401, forbidden 403, not_found 404, conflict 409, rate_limited 429, quota_exceeded 429 (a cap, until midnight UTC or a new conversation), connector_error 502 (the AI backend failed: retry), internal 500.
• Rate limits: each response has X-RateLimit-Limit and X-RateLimit-Remaining (requests a minute for the key); a 429 has Retry-After in seconds. Refused requests count too.
• Request ids: every response has X-Request-Id; quote it when asking for help.
• Size: JSON bodies up to 1 MB; file uploads up to 10 MB.
• Versioning: /api/v1 only grows: new endpoints, new fields, new enum values. Ignore fields you do not know. A breaking change would come as /api/v2, announced in the changelog.

Limits

Limit	Default	Setting
Requests a minute, per key	120	per key, default security.limits.apiRequestsPerKeyPerMinute
Active keys per site	50	security.limits.apiKeysPerSite
Messages per conversation	60	security.limits.messagesPerSession
Messages a day, whole site	500	security.limits.messagesPerSitePerDay (shared with the widget)
Message length	1000	security.limits.maxMessageLength

All are on Settings → Advanced; see Security.

From the command line

helppuff keys create "CRM sync" --scopes leads:read,leads:write --expires 90 --save HELPPUFF_CRM_KEY
helppuff keys list
helppuff keys revoke k7m3p9q2r4s8
helppuff api GET /leads
helppuff api POST /conversations --data '{"message": "Do you work weekends?"}'

helppuff api calls any endpoint with the project's admin key; see CLI reference.