Anthropic Messages API

Use the Anthropic-compatible surface when the selected AIXHUB model route supports Messages. Raw HTTP calls use the complete endpoint POST https://api.aixhub.org/v1/messages; Anthropic SDKs and clients usually receive the root Base URL https://api.aixhub.org and append /v1/messages themselves.

Create a dedicated AIXHUB API key before running the examples. Do not send an OpenAI Bearer header in place of x-api-key.

Bash request

read -rsp "AIXHUB API key: " AIXHUB_API_KEY && echo
export AIXHUB_API_KEY
export AIXHUB_MODEL="panel-anthropic-model-id"

curl --fail-with-body --silent --show-error \
  https://api.aixhub.org/v1/messages \
  --header "x-api-key: $AIXHUB_API_KEY" \
  --header "anthropic-version: 2023-06-01" \
  --header "content-type: application/json" \
  --data "{\"model\":\"$AIXHUB_MODEL\",\"max_tokens\":16,\"messages\":[{\"role\":\"user\",\"content\":\"Reply with OK only.\"}]}"

PowerShell request

$secureKey = Read-Host "AIXHUB API key" -AsSecureString
$env:AIXHUB_API_KEY = [System.Net.NetworkCredential]::new("", $secureKey).Password
$env:AIXHUB_MODEL = "panel-anthropic-model-id"

$headers = @{
  "x-api-key" = $env:AIXHUB_API_KEY
  "anthropic-version" = "2023-06-01"
}
$body = @{
  model = $env:AIXHUB_MODEL
  max_tokens = 16
  messages = @(@{ role = "user"; content = "Reply with OK only." })
} | ConvertTo-Json -Depth 4 -Compress

$response = Invoke-RestMethod `
  -Method Post `
  -Uri "https://api.aixhub.org/v1/messages" `
  -Headers $headers `
  -ContentType "application/json" `
  -Body $body
$response.content |
  Where-Object { $_.type -eq "text" } |
  Select-Object -ExpandProperty text

Replace the model stand-in with an exact Messages-compatible route. After testing in a disposable shell, run unset AIXHUB_API_KEY AIXHUB_MODEL in Bash. In PowerShell, remove the environment values and every object that holds the derived key:

Remove-Item Env:AIXHUB_API_KEY, Env:AIXHUB_MODEL -ErrorAction SilentlyContinue
Remove-Variable headers, body, response, secureKey -ErrorAction SilentlyContinue

Request fields

Start with the minimal fields. Add tools, images, prompt caching, or other optional features only after the route documentation confirms support and the basic request is visible in Usage.

Response example

The following JSON shows the stable shape to inspect. IDs and token counts are illustrative; do not hard-code them.

{
  "id": "msg_01Example",
  "type": "message",
  "role": "assistant",
  "model": "panel-anthropic-model-id",
  "content": [
    {
      "type": "text",
      "text": "OK"
    }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 12,
    "output_tokens": 1
  }
}

Read text blocks from content and inspect stop_reason before treating the output as complete. Preserve the HTTP status and any request ID header returned by the service when investigating a failure.

Streaming

Add "stream": true and disable curl buffering:

curl --no-buffer --fail-with-body --silent --show-error \
  https://api.aixhub.org/v1/messages \
  --header "x-api-key: $AIXHUB_API_KEY" \
  --header "anthropic-version: 2023-06-01" \
  --header "content-type: application/json" \
  --data "{\"model\":\"$AIXHUB_MODEL\",\"max_tokens\":16,\"stream\":true,\"messages\":[{\"role\":\"user\",\"content\":\"Reply with OK only.\"}]}"

A normal text stream follows an event sequence such as message_start, content_block_start, one or more content_block_delta events, content_block_stop, message_delta, and message_stop. ping events can appear between content events. Concatenate text only from text-delta content blocks in event order. Save the final message_delta.delta.stop_reason before message_stop.

Treat message_stop as the protocol end of the message, then inspect the saved stop_reason before accepting the output as complete. For this bounded OK test, expect end_turn; max_tokens means the output was truncated, while other reasons require application-specific handling. A stream can deliver an error event after the initial HTTP 200; in that case, keep the error, discard assumptions based on partial text, check Usage, and decide whether a new request is safe.

Errors and retries

A model ID or route problem is not always a 404. For any transient retry, allow at most three total attempts including the original, and honor Retry-After whenever the response actually returns it. Do not replay a 5xx, disconnect, or SSE error unless the application explicitly accepts duplicate output and charges, or can prove the prior generation was not accepted. See the error code reference for source classification and safe diagnostics.

Verification

The non-streaming example should print:

OK

For streaming, require message_stop and verify the final message_delta.delta.stop_reason is end_turn for this bounded test. Treat max_tokens as truncated rather than successful completion. Then open AIXHUB Usage and match the request time, selected model, and successful status. If Usage displays the protocol, confirm it is the Anthropic-compatible route.

Continue with the Anthropic Python SDK or configure Claude Code.

Official sources

• Anthropic Messages API
• Anthropic streaming Messages
• Anthropic API errors

Last verified: 2026-07-14