Anthropic
Use Claude for batch tasks like script generation. Same pattern as the OpenAI integration.
About the integration.
Bring your own Anthropic API key for batch tasks like KB writing, script generation, or evaluation grading. We default to our self-hosted AI for real-time chat and phone; Claude is a strong choice for offline tasks where latency matters less and quality matters more. Same abstraction layer as the OpenAI integration.
What you get on day one.
- ✓Bring your own Anthropic key. No markup, you pay Anthropic directly.
- ✓Ideal for batch tasks: KB writing, script generation, eval grading.
- ✓Claude Sonnet, Haiku, and Opus all supported.
- ✓Per-surface routing: pin Claude to one surface, leave the rest on our default AI.
- ✓Prompt caching supported (90% discount on cached prefix tokens).
- ✓Full prompt and response logs in your workspace.
Under the hood.
We expose an inference abstraction layer where you can register Anthropic alongside OpenAI and our default self-hosted AI. Add your Anthropic API key, pick which Claude models to enable, and optionally pin specific surfaces (e.g., the AI KB writer) to Claude. Claude is strongest on long-context comprehension and writing tasks where quality dominates. Real-time surfaces (chat, phone) stay on our default AI for latency reasons.
Set up in 6 steps.
- 01Sign in to console.anthropic.com and create an API key.
- 02Open Integrations, Anthropic in your message.com workspace.
- 03Paste the API key. Pick which Claude models to enable.
- 04Optional: pin batch surfaces (KB writer, eval grader, nightly summaries) to Claude under AI, Routing.
- 05Optional: enable prompt caching for repeat-prefix workloads to cut Claude cost by up to 90%.
- 06Save. Run a test inference from the AI Playground to verify the key works.
Questions people ask before they install.
Why would I use Claude when you ship your own AI?
Claude is strong for long-context writing and editing tasks (KB generation, script drafting). For real-time chat and phone, our self-hosted AI is faster and cheaper at our scale. Use Claude where quality dominates and latency is forgiving.
Does Anthropic train on my data?
No. Anthropic does not train on API data per their commercial terms.
Will Claude work for real-time chat?
It can, but Haiku is the only Claude fast enough, and we explicitly do not use Haiku for real-time customer-facing chat (it hallucinates more on grounded queries than our default AI does). Use Claude for batch tasks.
How is this different from the OpenAI integration?
Same abstraction, different provider. Bring whichever keys you have and route surfaces individually. Most customers use one or zero external providers; the integration is here for flexibility, not because it is required.
Other AI tools apps.
Try the Anthropic integration free.
14 days free, no credit card needed. Install the integration in under 5 minutes.
Start free trial →