One Chatbot. Any AI Model.
Most platforms pick a model vendor for you and bundle the cost into per-message pricing. Conferbot lets you choose from 21 providers per chatbot, bring your own API key, and change your mind later without rebuilding the flow.
Last updated September 2026 · Reviewed by the Conferbot team
How to pick a model provider
Four things actually decide this, and capability is rarely the one that matters most.
Cost. On a support bot answering ten thousand questions a month, the per-token gap between providers is the difference between a pilot and a budget line. Because you bring your own key, that saving is yours rather than ours.
Data residency. Some teams cannot send customer text to a US API. Selecting a European provider like Mistral, or a self-hosted model through Ollama, is a setting here rather than a migration.
Resilience. Every provider has outages. Configuring a second one means a vendor incident degrades your bot instead of ending the conversation.
Fit. Models differ more in tone and instruction-following than in raw capability. Running your own twenty questions through two providers settles it faster than any benchmark table.
United States
Providers developing and hosting models in United States.
OpenAI
Run your chatbot on OpenAI's GPT models
Anthropic Claude
Run your chatbot on Anthropic's Claude models
Google Gemini
Run your chatbot on Google's Gemini models
xAI (Grok)
Run your chatbot on xAI's Grok models
Perplexity
Run your chatbot on Perplexity's search-grounded models
Groq
Run open-weight models at very low latency on Groq
Cerebras
Run open-weight models on Cerebras wafer-scale inference
SambaNova
Run open-weight models on SambaNova's inference cloud
Fireworks AI
Run a wide catalogue of open-weight models on Fireworks
Together AI
Run a large catalogue of open-weight models on Together AI
DeepInfra
Run open-weight models cheaply on DeepInfra
OpenRouter
Reach hundreds of models through a single OpenRouter key
China
Providers developing and hosting models in China.
DeepSeek
Run your chatbot on DeepSeek's reasoning and chat models
Qwen (Alibaba)
Run your chatbot on Alibaba's Qwen models
Z.ai (Zhipu)
Run your chatbot on Zhipu's GLM models
Moonshot (Kimi)
Run your chatbot on Moonshot's long-context Kimi models
MiniMax
Run your chatbot on MiniMax's multimodal models
France
Providers developing and hosting models in France.
Canada
Providers developing and hosting models in Canada.
Singapore
Providers developing and hosting models in Singapore.
Self-hosted
Run the model on hardware you control. No customer text reaches a third-party model vendor, while the widget, inbox, handover and analytics stay fully managed.
Provider comparison
Every provider speaks an OpenAI-compatible endpoint, so switching is a configuration change rather than an integration project.
| Provider | Jurisdiction | Open weights | Best for |
|---|---|---|---|
| OpenAI | United States | No | Teams that want the safest default |
| Anthropic Claude | United States | No | Regulated or sensitive support conversations |
| Google Gemini | United States | No | High-volume bots where cost per conversation dominates |
| DeepSeek | China | Yes | High-volume customer support |
| xAI (Grok) | United States | No | News, markets and other fast-moving subjects |
| Mistral AI | European Union | Yes | EU businesses with GDPR or data-residency constraints |
| Cohere | Canada | No | Large internal knowledge bases |
| Perplexity | United States | No | Research and comparison assistants |
| Groq | United States | Yes | High-traffic website widgets where responsiveness matters |
| Cerebras | United States | Yes | Latency-sensitive widgets |
| SambaNova | United States | Yes | Enterprise teams standardising on open models |
| Fireworks AI | United States | Yes | Teams comparing several open models |
| Together AI | United States | Yes | Evaluating open models before committing |
| DeepInfra | United States | Yes | Cost-minimising high-volume bots |
| Novita AI | Singapore | Yes | Asia-Pacific businesses |
| OpenRouter | Routed | No | Teams testing many models quickly |
| Qwen (Alibaba) | China | Yes | Chinese-language support bots |
| Z.ai (Zhipu) | China | Yes | Bilingual Chinese and English support |
| Moonshot (Kimi) | China | Yes | Bots answering from long manuals or policy documents |
| MiniMax | China | Yes | Bots handling voice or media input |
| Ollama (self-hosted) | Your own infrastructure | Yes | Healthcare, legal, government and defence |
What stays the same whichever model you pick
The provider is a setting, not an architecture. Everything below is identical across all 21: the visual flow builder, the knowledge base, live chat handover, analytics, the template library, and deployment to your website, WhatsApp, Telegram, Discord, Slack and every other channel.
Start on the free plan with 600 conversations a month - provider choice is not gated by plan.
Choosing a model provider
How many AI model providers does Conferbot support?
Conferbot can route a chatbot to 21 providers: OpenAI, Anthropic Claude, Google Gemini, DeepSeek, xAI (Grok), Mistral, Cohere, Perplexity, Qwen, Z.ai, Moonshot (Kimi), MiniMax, Groq, Cerebras, SambaNova, Fireworks, Together, DeepInfra, Novita, OpenRouter, and self-hosted models through Ollama. You supply your own API key for whichever you choose.
Can I switch AI models without rebuilding my chatbot?
Yes. The provider is a setting on the chatbot, not part of the flow. Your conversation design, knowledge base, transcripts, widget and embed code are all provider-agnostic, so switching from OpenAI to DeepSeek or from Claude to Mistral takes one change and no rebuild. You can also run different providers on different chatbots in the same workspace.
Do I need my own API key, and does Conferbot mark up tokens?
You bring your own key, and there is no markup. Token usage bills directly from you to the provider at their published rates, which means choosing a cheaper model reduces your cost rather than the vendor's. Keys are encrypted at rest, scoped to one workspace, never exposed to the browser and never shared between providers.
What happens if my AI provider goes down?
Configure a second provider and conversations continue on it instead of failing in front of the visitor. Without a fallback, the AI nodes error while the scripted parts of your flow and the human handover keep working. A fallback costs nothing until it is used, which is the reason to set it up before you need it.
Can I run an AI chatbot without sending data to a US provider?
Yes, in two ways. Select a non-US provider such as Mistral in France, or avoid external APIs entirely by serving an open-weight model on your own hardware and pointing Conferbot at a self-hosted Ollama endpoint. In both cases the flow, widget and inbox are unchanged.
Is model choice limited by plan?
No. Provider choice is not gated by plan - the free plan can use any supported provider. What differs by plan is conversation volume and features such as webhooks, API access and removing branding.
Go deeper on AI models
How model choice affects cost, accuracy and tone, and the terminology behind the decision.
AI Model Providers FAQ
Everything you need to know about chatbots for ai model providers.
Continue Exploring
Explore features, connect third-party tools, and browse ready-made templates.
Deep-dive pillar guides, real use cases, and the chatbot & AI glossary.