Who a Claude chatbot suits, and who it does not
Claude is the family of models made by Anthropic. A Claude chatbot is a sensible choice when your agent has a lot of rules to follow: a tone of voice, things it must never promise, a returns policy it must quote exactly. Our Anthropic key guide singles out strong instruction following, which matters most with strict knowledge mode and a firm prompt. It is not a guarantee, so test awkward questions yourself.
It is a weaker fit in three situations:
- You only have a Claude.ai subscription. A Pro or Max plan for the chat app does not include API access. You need a separate Anthropic Console account with its own billing.
- You want a free allowance. Anthropic’s API is typically prepaid, and with no credit requests fail even though the key is valid. The Gemini free tier is the usual route for free testing.
- You want a single invoice. BYOK means one more account to keep funded; see the BYOK overview.
Which Claude model to pick, and what it costs
The table lists every Claude model in our catalog with its list price and an estimate of what 1,000 conversations would cost. Anthropic’s prices in our catalog were confirmed against its documentation on 2026-10-07, so no row is marked as unverified, but prices change and the Anthropic Console is always authoritative.
| Model | Input, per 1M tokens | Output, per 1M tokens | Estimate per 1,000 conversations |
|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $179 |
| Claude Opus 5.5 | $4.00 | $20.00 | $71.76 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $35.88 |
| Claude Haiku 5.5 | $0.10 | $0.50 | $1.79 |
Estimates, not quotes: list prices from our model catalog in US dollars per million tokens, for 6 visitor messages per conversation, 1,500 tokens of prompt and knowledge sent every turn, and 180-token replies, with no prompt caching. Your provider's bill is authoritative.
Read the spread rather than any single row. Claude Haiku 5.5 comes out at about $1.79 per 1,000 conversations and Claude Sonnet 5.5 at about $35.88, a large multiple. For an agent that answers from pasted facts, a small model is a reasonable first test; move up only when the playground shows real mistakes, such as dropped constraints. Claude Opus 5.5 and Claude Fable 5.1 are for the rare agent that needs deeper reasoning. Pick from the dashboard list rather than typing an identifier, since models are retired from time to time, and see how to choose an LLM for website chat.
Set up your Claude chatbot, step by step
Allow about ten minutes. If an Anthropic menu has moved, follow the Connect Anthropic (Claude) guide and Anthropic’s own help.
Add credit in the Anthropic Console
Sign in at console.anthropic.com (not Claude.ai), open billing and buy prepaid credit. Without a balance, a valid key still fails.
Create a dedicated key
Open API keys and create a key named “SiteTrellis production” in the workspace you want to bill. Copy it at once; it starts with
sk-ant-and is shown only once.Add the key in SiteTrellis
Under AI providers choose Anthropic, label the key and paste it. An automatic test should turn the status to ok; the AI keys overview covers testing and rotation.
Choose the model
In the agent’s model section add a step, select this key and pick a Claude model from the list. Begin with the small, low-cost one.
Set a spend limit at Anthropic
In the Console’s limits settings set a monthly limit, per workspace if you use them, at an amount you could stand to lose.
Test in the playground
Ask your ten most common questions, then try to break the agent. Playground chats cost a little but are not counted.
Publish and add the script
Publish, then paste the one-line snippet on your site. The quickstart covers it.
Keeping the Anthropic bill predictable
Prepaid billing gives you a natural ceiling: the credit you load is the most a runaway chat can spend.
- If you enable auto-reload, pair it with a monthly limit, or the safety net disappears.
- Lower the history window (default 20 messages) and max reply tokens if conversations run long; every turn resends the history.
- Keep pasted knowledge tight, because it is sent with every message.
- Set a monthly message cap on the agent for a second ceiling inside SiteTrellis.
To model your own traffic, use the AI chatbot cost calculator.
An example fallback chain for a Claude chatbot
Anthropic’s tier-based rate limits grow with your paid history, and an empty prepaid balance stops the key outright. A second step on a different provider covers both.
| Step | Key and model | Why |
|---|---|---|
| 1 | Anthropic key, Claude Haiku 5.5 | Fast and low cost for most questions, with Claude’s instruction following. |
| 2 | OpenAI key, GPT-6 Luna | A different company, so a 429, a 529 overload, an Anthropic incident or an empty balance does not reach the visitor. |
The chain moves on after a network error, a timeout, a 429, a 5xx or overloaded reply, an invalid key or an empty reply, but not after a content-policy refusal or a bad request. Free allows 2 steps and Starter 3. The reasoning is in why one provider is a single point of failure.
Claude and Anthropic gotchas
| You see | What it means | What to do |
|---|---|---|
| 401 | Wrong, revoked or whitespace-padded key. | Create a new key, paste it cleanly and rotate it in. |
| 403 | The key’s workspace cannot use that model or feature. | Check workspace permissions or choose another model. |
| 404 | The model is retired or not enabled for you. | Choose a current model from the dashboard list. |
| “Credit balance is too low” | The prepaid balance is empty. | Add credit in the billing area, then press Test. |
| 429 | Requests faster than your tier allows. | Wait, raise your tier, or rely on a fallback step. |
| 500 or 529 | Anthropic is busy or has an incident. | Transient. A fallback step covers it. |
One quieter point: our catalog marks the current Claude models as not taking a temperature setting, so shape tone through the prompt rather than the slider. More fixes are on the troubleshooting page.
Data handling and honest limits
Your visitors’ messages, your prompt and your knowledge text go to Anthropic under your account’s terms. Anthropic publishes how it treats API data, including retention and training; read the current terms and reflect them in your privacy policy. How SiteTrellis protects your key is in Security and privacy.
Limits: knowledge is pasted text with no crawler, replies arrive whole rather than streaming, a model can still state something wrong, and you must keep the balance funded. Route high-stakes questions to a person with human handoff.