Skip to main content

​
AI Provider Configuration

Configure your preferred AI provider to power ChatterMate’s intelligent customer support capabilities.
AI Provider Configuration

​
Available Providers

ChatterMate supports multiple AI providers, giving you the flexibility to choose the one that best fits your needs:

OpenAI

GPT-4.1, GPT-4o, and o-series models

Anthropic (Claude)

Claude Opus, Sonnet, and Haiku

Google Gemini

Gemini 2.5 Pro / Flash

Mistral

Mistral Large, Medium, and Small

xAI (Grok)

Grok 4 and Grok 3

DeepSeek

DeepSeek Chat and Reasoner

Groq

Ultra-fast inference (GPT-OSS, Llama)

OpenAI-compatible

Self-hosted or gateway endpoints — Ollama, vLLM, llama.cpp, OpenRouter, LiteLLM
Every provider is bring-your-own-key — you supply your own API key, encrypted at rest. When you pick a provider, a “Get your API key” link points to that provider’s console. The exception is OpenAI-compatible, which has no hosted console: you point ChatterMate at an endpoint you control or subscribe to, so the key comes from that endpoint.
Want to keep inference on your own hardware? Pick OpenAI-compatible and give ChatterMate the base URL of your own server. See Self-hosted and gateway models for worked Ollama, vLLM, llama.cpp, OpenRouter, and LiteLLM setups.

​
Configuration Fields

​
AI Provider

Select your provider from the grid. The seven hosted providers all support tool calling and structured output, so any of them works with ChatterMate’s agents, knowledge search, and workflows:
  • OpenAI (Recommended): Best-in-class models with consistent performance
  • Anthropic (Claude): Strong reasoning and analysis
  • Google Gemini: Large context windows, great price/performance
  • Mistral / xAI (Grok) / DeepSeek / Groq: Cost and speed options
  • OpenAI-compatible: Any endpoint that speaks the OpenAI chat-completions API — your own server (Ollama, vLLM, llama.cpp) or a gateway (OpenRouter, LiteLLM). This is the option to pick when the model should run on your own hardware.
Features that depend on structured output — automatic lead capture, ending a resolved chat, and human handoff — are most consistent on OpenAI and Anthropic models. Smaller open models (e.g. Groq’s GPT-OSS, some Grok variants) handle these well most of the time but can occasionally miss a step. If you find a model isn’t capturing leads or closing chats reliably, switch to a stronger model like GPT-4.1 or Claude.This applies with much more force to OpenAI-compatible endpoints, because the model behind them is whatever you chose to run. Many small local models have weak tool calling or none at all, and a server that doesn’t implement tools or response_format will make knowledge search, lead capture, chat-ending, and handoff fail rather than degrade quietly. Pick a model whose card explicitly claims tool-calling support, and test those flows before you go live.

​
Base URL

Only shown for the OpenAI-compatible provider — the other seven have no base URL field, because ChatterMate already knows where to reach them. Enter the root of your endpoint’s OpenAI-compatible API, including the version segment. ChatterMate appends the rest of the path itself:
Trailing slashes and a trailing /chat/completions are trimmed for you, so pasting the full completions URL works too. The URL must be reachable from the ChatterMate backend, not from your browser — see Self-hosted and gateway models for the Docker networking details.

​
Model Name

Pick a suggested model from the dropdown, or choose “Custom model ID” to type any model your provider supports (handy when a provider ships a new model before this list updates).
OpenAI-compatible has no suggested models. The model list belongs to your endpoint, not to ChatterMate, so the dropdown offers only “Custom model ID…” — select it and type the exact model ID your server reports. ollama list, or curl <base-url>/models -H "Authorization: Bearer <key>", tells you what to enter.
Suggested models for the hosted providers:
Model IDs change over time as providers release and retire models. If a suggested ID stops working, use Custom model ID and paste the current identifier from your provider’s docs.

​
API Key

Your provider’s API key for authentication. Security measures:
  • Keys are encrypted before storage
  • Keys are never logged or exposed in plaintext
  • Access is restricted to authorized systems only
  • Regular key rotation is supported
  • Compliance with SOC 2 standards
Never share your API keys or commit them to version control. ChatterMate encrypts and stores them securely.
An API key is always required, including for OpenAI-compatible endpoints. If your server doesn’t check authentication — the default for Ollama and llama-server — enter any non-empty placeholder (ollama is the conventional one). The value is sent as a bearer token and ignored by a server that isn’t looking at it.
Saving the form runs a live test message against the provider with the key, model ID, and base URL you entered. A save that fails with “Invalid API key” on an OpenAI-compatible endpoint usually means the backend couldn’t reach the URL, or the model ID doesn’t exist on that server — not that the key itself is wrong.

​
Best Practices

  1. Provider Selection
    • Start with OpenAI’s GPT-4.1 / GPT-4o for best overall performance
    • Consider Anthropic’s Claude for complex reasoning tasks
    • Use Groq (GPT-OSS / Llama) when you want the fastest, lowest-cost inference
    • Use OpenAI-compatible when prompts and customer messages must not leave your infrastructure
  2. Model Choice
    • Balance between capability and cost
    • Test different models before production
    • Monitor usage and adjust as needed
    • On a self-hosted model, verify tool calling works before going live — see Self-hosted and gateway models
  3. Security
    • Use environment-specific API keys
    • Rotate keys periodically
    • Monitor usage for anomalies

​
What’s Next?

After configuring your AI provider:
  1. Test the configuration with sample queries
  2. Customize the AI agent’s behavior
  3. Add domain knowledge to improve responses
  4. Set up human handoff rules

Customize AI Agent

Next: Learn how to customize your AI agent’s behavior

Self-hosted and gateway models

Run the model on your own hardware with an OpenAI-compatible endpoint