Skip to main content

Anthropic Provider

💡TL;DR

Anthropic uses the native Messages API transport, not the OpenAI-compatible transport. The current preset defaults to claude-3-5-sonnet-20240620, https://api.anthropic.com, and temperature: 0.5. Treat it as a separate protocol surface when debugging.

This is part of the Obsidian AI Knowledge Management Guide.

Setup​

Create an Anthropic API key, add an Anthropic provider profile, and leave the Base URL on the native endpoint unless you are routing through a dedicated gateway profile. Choose the Claude model ID you want to use in the model field; the default keeps the existing product baseline stable.

Anthropic is a good candidate for high-quality summarization, research synthesis, and concept extraction. If you mix it with cheaper providers, set it only on the tasks that benefit from its output quality.

Endpoint And Authentication​

FieldCurrent preset
Transportanthropic
API keyRequired
Base URLhttps://api.anthropic.com
Default modelclaude-3-5-sonnet-20240620
Temperature0.5
Connection testChat-only

The runtime sends requests to /v1/messages with x-api-key and anthropic-version: 2023-06-01. Do not configure an OpenAI-style /v1/chat/completions endpoint under this preset; use an OpenAI-compatible gateway preset instead.

Model Discovery​

Anthropic uses the anthropic-models discovery mode in the provider registry. If your key or account cannot list models, keep a known model ID in the profile and validate through the connection test.

The transport is not interchangeable with OpenAI-compatible providers. A Claude model routed through OpenRouter, LiteLLM, or another gateway belongs in that gateway profile, not in the native Anthropic profile.

Troubleshooting​

  • 401 or 403: check key scope, workspace access, billing, and whether the key belongs to the expected Anthropic workspace.
  • 404: the Base URL is wrong or the selected model ID is not available to the account.
  • Gateway-style Claude model IDs fail: move the profile to the matching gateway provider because native Anthropic expects native model names.
  • Long requests fail late: lower per-task model output size or use a model with a larger known output-token ceiling.

When To Use​

Use Anthropic when quality on synthesis, research notes, or structured writing matters more than raw cost. Do not use it as a generic OpenAI-compatible fallback; the separate transport is the point, and mixing the protocols makes diagnosis harder.

Next Steps​