Skip to main content
ReplyBubble uses Bring Your Own Key (BYOK) — you connect your own AI account and pay your provider directly. ReplyBubble charges a flat platform fee and never marks up AI costs.
Time estimate: ~5 minutes

​
What is BYOK?

Most chatbot platforms bundle AI costs into their pricing and charge you per message or per conversation. ReplyBubble works differently:
  1. You create an account with an AI provider (OpenAI or Anthropic)
  2. You generate an API key and paste it into ReplyBubble
  3. When a visitor asks a question, ReplyBubble sends it to your AI provider and returns the answer
  4. The AI provider bills you directly at their standard rates
Why this matters: You get full transparency into AI costs, no markup, and the freedom to switch providers or models at any time.
OpenAI is the most popular choice. Their GPT-4o-mini model offers a great balance of quality and cost for most use cases.

​
Get your API key

  1. Go to platform.openai.com/api-keys
  2. Sign in or create an account
  3. Click Create new secret key
  4. Name it “ReplyBubble” (helps you identify it later)
  5. Click Create secret key
  6. Copy the key immediately — it starts with sk-proj- and won’t be shown again
Important: Add billing to your OpenAI account at platform.openai.com/account/billing. Without billing, API calls will fail.

​
Connect to ReplyBubble

  1. In your ReplyBubble dashboard, go to Settings > AI Provider
  2. Select OpenAI
  3. Paste your API key
  4. Click Save

​
Setting up Anthropic

Anthropic makes the Claude family of models, which are known for careful, nuanced responses.

​
Get your API key

  1. Go to console.anthropic.com
  2. Sign in or create an account
  3. Go to API Keys
  4. Click Create Key
  5. Name it “ReplyBubble”
  6. Copy the key — it starts with sk-ant-

​
Connect to ReplyBubble

  1. In your ReplyBubble dashboard, go to Settings > AI Provider
  2. Select Anthropic
  3. Paste your API key
  4. Click Save

​
Choosing a model

After connecting your provider, select which AI model to use. Here’s a simple comparison:

​
OpenAI models

​
Anthropic models

Note: Anthropic updates model versions regularly. Your ReplyBubble dashboard shows the specific models available — pick from there.
Recommendation: Start with GPT-4o Mini (OpenAI) or Claude Haiku 4.5 (Anthropic) — these are the defaults. They handle most support and sales conversations well at the lowest cost. Upgrade to a more capable model if you find answers need more depth.

​
Understanding AI costs

AI providers charge per token (roughly 1 token = 0.75 words). A typical chatbot conversation uses 500–2,000 tokens.

​
Realistic cost estimates

These are estimates. Actual costs depend on conversation length and complexity.

​
How ReplyBubble keeps costs low

ReplyBubble automatically reduces your AI spend in several ways:
  • Response caching — repeat questions get instant answers without calling the AI
  • Spam filtering — spam and unserious messages are blocked before using any AI credits
  • Efficient knowledge base search — only relevant content is sent to the AI, not your entire knowledge base
  • Offline mode — when the AI is unavailable, visitors get pre-written answer flows at zero AI cost (see Settings)
  • Daily token quotas — set a maximum daily AI usage in Dashboard > Settings to prevent surprise bills

​
Switching providers

You can switch between OpenAI and Anthropic at any time:
  1. Go to Settings > AI Provider
  2. Select the new provider
  3. Paste the new API key
  4. Click Save
Your knowledge base, conversation history, and all settings are preserved. Only the AI model powering the responses changes.

​
Troubleshooting

​
”API key invalid” error

  • Make sure you copied the full key (no extra spaces)
  • Verify billing is set up on your AI provider account
  • Check that the key hasn’t been revoked in your provider’s dashboard

​
Bot not responding

  • Confirm your API key is active and has billing enabled
  • Check your AI provider’s status page for outages
  • Verify you haven’t hit your provider’s rate limits or spending caps

​
Responses are slow

  • Try switching to a faster model (GPT-4o-mini or Claude Haiku)
  • Check your AI provider’s status page — response times can vary during peak usage
For more help, see Troubleshooting.