Anthropic-compatible servers

Expose a server that speaks the Anthropic Messages API.

Noket forwards both OpenAI- and Anthropic-shaped requests. Choose the protocol per upstream; a server that speaks the Anthropic Messages API uses the anthropic protocol.

noket upstream add \
  --name local-anthropic \
  --protocol anthropic \
  --url http://127.0.0.1:8080 \
  --model my-model

What changes with the `anthropic` protocol

	`openai`	`anthropic`
Path	`/v1/chat/completions` and related	`/v1/messages`
Key header	`Authorization: Bearer …`	`x-api-key: …`
Model discovery	`GET {base}/v1/models`	`GET {base}/v1/models`

Probing for models uses GET {base}/v1/models regardless of protocol, because that is the endpoint servers commonly expose. A server that does not implement it will fail the probe; save the upstream anyway and enter model ids by hand.

Custom headers

Some gateways need extra headers, for example a routing or organization header. There is no command-line flag for them; add them in config.toml:

[[upstreams]]
name = "gateway"
protocol = "anthropic"
base_url = "https://gateway.internal"
headers = { "X-Route" = "team-a" }
models = ["claude-model"]

See the configuration reference for every key.

When to use which protocol

Match the server, not the client. If the server implements /v1/messages, use anthropic; if it implements /v1/chat/completions, use openai. A proxy that translates between the two should be configured with the protocol it exposes, and Noket will not translate anything on its own.

Next