Providers

Providers

HelpPuff's assistant has two parts you choose separately:

• The model (model in helppuff.json): who writes the answers.
• The knowledge (knowledge.retrieval): what the answers come from.

The assistant around them stays the same whichever you pick: the prompt and its rules, the tools (callbacks, jobs, live chat hand-over, opening hours), citations, guardrails, history, the widget, dashboard, leads and Webhooks. Both are changed with the CLI and a deploy, never in the dashboard.

Model (model.provider)	What writes the answers	Needs
workers-ai (default)	Cloudflare Workers AI (GLM-4.7 Flash by default). See [[Workers AI	Provider-Workers-AI]]
openai-compatible	Any OpenAI-compatible API with tools, with presets for DeepInfra, OpenRouter, DeepSeek, Groq, Together, Mistral, Fireworks, the Vercel AI Gateway and the Cloudflare AI Gateway	that provider's key
openai	OpenAI	OPENAI_API_KEY
gemini	Google Gemini	GEMINI_API_KEY
anthropic	Anthropic Claude	ANTHROPIC_API_KEY
custom	Your own code: a TypeScript file with one chat function. See [[Custom model	Custom-Model]]

Knowledge (knowledge.retrieval.type)	What answers come from	Filled by
helppuff (default)	HelpPuff's own knowledge base: your site and files, in Vectorize + D1 on your account. See [[Knowledge base	Knowledge-Base]]
none	Nothing but the prompt and the business details	n/a
ai-search	Cloudflare AI Search	AI Search's crawl, or your files on deploy
openai-vector-store	An OpenAI vector store	you, in OpenAI
http	Your own search endpoint	you
custom	Your own code: a TypeScript file with one search function. See [[Custom knowledge base	Custom-Knowledge-Base]]

The full list of models, presets and how to switch: Models and providers.

Backends that run the whole conversation

A few backends run the conversation themselves, so they take no model or knowledge.retrieval; set backend instead:

backend.type	What answers	Needs
[[retell	Provider-Retell]]	A Retell chat agent (it can search HelpPuff's knowledge base through a custom function)
[[http	Provider-Your-Own-API]]	Your own API, speaking the HelpPuff backend protocol
[[openai	Provider-OpenAI]]	OpenAI's Responses API with a stored prompt, a vector store HelpPuff fills for you, or another base URL
[[gemini	Provider-Gemini]]	Gemini with File Search
echo	Nothing: echoes what it is sent and demonstrates every widget feature	nothing (development only)

Choosing

• Start with the defaults: Workers AI and HelpPuff's knowledge base. Free, only a Cloudflare login, and it answers only from your own site and files, saying "not sure" (and offering a callback) rather than guessing.
• Want a particular model? Change only model. Your crawl, files, business details and dashboard stay as they are.
• Your knowledge lives elsewhere? Change only knowledge.retrieval: AI Search, an OpenAI vector store, your own search over HTTP, or your own code.
• Already have a whole agent? Use retell, or your own API (http).

Switching

helppuff model set openai-compatible --preset deepinfra --model deepseek-ai/DeepSeek-V3.1
helppuff secret set DEEPINFRA_API_KEY
helppuff deploy
helppuff model test "Do you do emergency callouts?"

Conversations, leads, settings and the knowledge base are kept. The prompt (prompt.md) carries over to every model; Retell, an OpenAI stored prompt and your own API in helppuff mode keep theirs.

Costs and limits

Workers AI is counted in neurons against the daily budget (model.budget.dailyNeurons, 9,000 by default, under the free 10,000). Other providers bill you directly at their own prices: cap them with the daily message limit, security.limits.messagesPerSitePerDay. See Costs and limits.

What every backend gets

• The same widget: rich messages (option chips, cards, link lists, forms), streaming where the backend streams, ratings, lead form, flows.
• The same dashboard, leads and Webhooks.
• The same limits and security: origin allowlist, signed sessions, rate limits, the daily cap, optional Turnstile.