Aeion AI + Anthropic Claude
Configure Anthropic Claude as a provider in Aeion AI. Covers the current Claude family (Opus / Sonnet 4.6, Opus / Sonnet 4.5, Haiku 4.5), vision input, and native tool use through the direct Anthropic API.
Setup
- Go to
/admin/ai/providersand create a new AI Provider record.
Model Family — When to Use Which
Claude Sonnet / Opus 4.6 are the current flagship models, alongside the still-active 4.5 generation.
Cost Tracking
Every Claude call routed through Aeion is logged with token usage and latency, so per-tenant and per-model spend is visible without instrumenting anything yourself.
Deployment
Aeion's Anthropic adapter calls the direct Anthropic API (api.anthropic.com). AWS Bedrock and Google Vertex hosting for Claude are not currently wired up as separate Aeion provider tenancies — if you need Bedrock or Vertex today, front it with your own OpenAI/Anthropic-compatible gateway and register that endpoint as a custom provider.
Vision + Tool Use
Claude's multi-modal surface.
Common Gotchas
- Rate limits tighter than OpenAI — pair with
groqorgooglein a fallback chain for spike resilience.
FAQ
Opus for the hardest reasoning + agentic loops with many tool calls. Sonnet for the default — handles the large majority of production workloads at 5× lower cost than Opus. Haiku for bulk classification, summarization, and short answers — the cheapest tier by a wide margin.
Yes — pass a schema and Aeion uses Anthropic's structured-output capability, implemented behind the scenes as tool-calling with a schema-defined output tool.
All current Claude models support 200K input tokens. Opus 4.5 and Sonnet 4.5 additionally expose a 1M-token extended-context preview.
Not yet as a first-class option — Aeion's Anthropic adapter doesn't currently set cache-control markers on requests, so repeat-prompt discounting isn't automatic. Token usage and cost are still tracked per call regardless.