Native Anthropic streaming

Use the native Anthropic route when your application depends on Anthropic’s request shape, SDK helpers, or named SSE events. The native base path is:

https://api.inquantum.ai/anthropic/v1/*

Native passthrough uses two credentials:

Planck-Auth authenticates the request to Planck.
x-api-key authenticates the request to Anthropic.

For provider-independent routing with an Anthropic-style request, use the mapped POST /v1/messages route instead.

Stream with the Anthropic SDK

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
  baseURL: "https://api.inquantum.ai/anthropic",
  defaultHeaders: {
    "Planck-Auth": `Bearer ${process.env.PLANCK_API_KEY}`,
  },
});

const stream = client.messages
  .stream({
    model: "claude-4.5-sonnet",
    max_tokens: 512,
    messages: [{ role: "user", content: "Explain vector search simply." }],
  })
  .on("text", (text) => process.stdout.write(text));

const message = await stream.finalMessage();
console.log(`\n${message.usage.output_tokens} output tokens`);

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["ANTHROPIC_API_KEY"],
    base_url="https://api.inquantum.ai/anthropic",
    default_headers={
        "Planck-Auth": f"Bearer {os.environ['PLANCK_API_KEY']}"
    },
)

with client.messages.stream(
    model="claude-4.5-sonnet",
    max_tokens=512,
    messages=[{"role": "user", "content": "Explain vector search simply."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

    message = stream.get_final_message()
    print(f"\n{message.usage.output_tokens} output tokens")

curl -N https://api.inquantum.ai/anthropic/v1/messages \
  -H "Planck-Auth: Bearer $PLANCK_API_KEY" \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-4.5-sonnet",
    "max_tokens": 512,
    "messages": [
      {"role": "user", "content": "Explain vector search simply."}
    ],
    "stream": true
  }'

Event lifecycle

Native Anthropic streams use named SSE events rather than OpenAI chat.completion.chunk objects.

message_start opens the message and includes initial usage.
content_block_start opens a text, thinking, or tool-use block.
One or more content_block_delta events add content to that block.
content_block_stop closes the block.
message_delta updates message-level fields such as the stop reason and cumulative usage.
message_stop closes the message.

ping events can appear anywhere between those events.

Native SSE response

event: message_start
data: {"type":"message_start","message":{"id":"msg_123","type":"message","role":"assistant","content":[],"model":"claude-4.5-sonnet","stop_reason":null,"usage":{"input_tokens":14,"output_tokens":1}}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}

event: ping
data: {"type":"ping"}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Vector search"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" finds items with similar meaning."}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"output_tokens":12}}

event: message_stop
data: {"type":"message_stop"}

Text, thinking, and tool deltas

Inspect the delta type instead of assuming every content event contains text.

Delta type	Field to accumulate	When to process
`text_delta`	`delta.text`	Render immediately.
`thinking_delta`	`delta.thinking`	Preserve only when requested and allowed.
`signature_delta`	`delta.signature`	Preserve with its thinking block.
`input_json_delta`	`delta.partial_json`	Concatenate, then parse after `content_block_stop`.

Tool arguments in input_json_delta are partial JSON strings. They follow the same rule as streamed OpenAI tool calls: accumulate first, parse once the block is complete.

Errors and forward compatibility

An error can arrive after the HTTP stream has started:

Error event

event: error
data: {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}

Handle error events even when the HTTP response began with 200. Ignore unknown event types safely; Anthropic can add event types over time without changing the API version.

Choose the right route

Requirement	Recommended route
OpenAI SDK or provider-independent routing	`POST /v1/chat/completions`
Anthropic-style request with gateway mapping	`POST /v1/messages`
Anthropic SDK and native event semantics	`/anthropic/v1/messages`
Other native Anthropic endpoints	`/anthropic/v1/*`

Native Anthropic reference

Review authentication and native path behavior.

OpenAI-compatible streaming

Stream across supported providers with Chat Completions.

Getting Started

AI Gateway

Observability & Analytics

Prompt Management

Async Logging

Custom Logs

Native Anthropic streaming

Stream with the Anthropic SDK

Event lifecycle

Text, thinking, and tool deltas

Errors and forward compatibility

Choose the right route

Native Anthropic reference

OpenAI-compatible streaming

​Stream with the Anthropic SDK

​Event lifecycle

​Text, thinking, and tool deltas

​Errors and forward compatibility

​Choose the right route

Native Anthropic reference

OpenAI-compatible streaming

Stream with the Anthropic SDK

Event lifecycle

Text, thinking, and tool deltas

Errors and forward compatibility

Choose the right route