One base URL. Three dialects.

Authentication

Create an API key in your dashboard. Keys start with bcl_ and are shown once. Send one of:

Authorization: Bearer bcl_...
# or, for Anthropic clients:
x-api-key: bcl_...
anthropic-version: 2023-06-01

The key belongs to your account. Requests are billed to your balance using the model's base price and your account discount.

Models

Use the exact model slug from the catalog. The same slug works across all three APIs. List them with:

curl https://backchannel.lol/v1/models \
  -H "Authorization: Bearer $BACKCHANNEL_KEY"

Chat Completions

curl https://backchannel.lol/v1/chat/completions \
  -H "Authorization: Bearer $BACKCHANNEL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-model",
    "messages": [
      { "role": "system", "content": "Be concise." },
      { "role": "user", "content": "What is a backchannel?" }
    ],
    "stream": true
  }'

Responses

Supports string or item-array input, instructions, function tools, and streaming response events.

curl https://backchannel.lol/v1/responses \
  -H "Authorization: Bearer $BACKCHANNEL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-model",
    "instructions": "Be concise.",
    "input": [
      { "role": "user", "content": "What is a backchannel?" }
    ],
    "stream": true
  }'

Anthropic Messages

Supports system blocks, text/image/tool_use/tool_result content, stop sequences, and the Messages stream. max_tokens defaults to 4096 if omitted.

curl https://backchannel.lol/v1/messages \
  -H "x-api-key: $BACKCHANNEL_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-model",
    "max_tokens": 512,
    "system": "Be concise.",
    "messages": [
      { "role": "user", "content": "What is a backchannel?" }
    ],
    "stream": true
  }'

SDK setup

Use the vendor SDKs by changing only the base URL and key. All translation happens at the proxy.

from openai import OpenAI

client = OpenAI(
    api_key="bcl_...",
    base_url="https://backchannel.lol/v1",
)

resp = client.chat.completions.create(
    model="your-model",
    messages=[{"role": "user", "content": "hello"}],
)
print(resp.choices[0].message.content)

from anthropic import Anthropic

client = Anthropic(
    api_key="bcl_...",
    base_url="https://backchannel.lol",
)

msg = client.messages.create(
    model="your-model",
    max_tokens=512,
    messages=[{"role": "user", "content": "hello"}],
)
print(msg.content[0].text)

const res = await fetch("https://backchannel.lol/v1/responses", {
  method: "POST",
  headers: {
    "authorization": `Bearer ${process.env.BACKCHANNEL_KEY}`,
    "content-type": "application/json",
  },
  body: JSON.stringify({
    model: "your-model",
    input: "hello",
  }),
});

const data = await res.json();
console.log(data.output);

Translation

Each model/provider route declares a native API: chat, responses, or messages. Any incoming dialect is translated to the route's native dialect, and the response is translated back. Streaming uses a canonical event stream, so text, tool calls, usage, and stop reasons survive across formats.

Pricing & billing

Prices are per million input and output tokens. New accounts start at 90% off. Your discount is set by lifetime top-ups and locked for the life of the account; admin overrides can supersede it. Models can set a separate cached-input price; cached reads use that instead of the normal input price.

A request costs:

non_cached_input = input_tokens - cached_input_tokens
cached_price = model.cached_input_price ?? input_price
base = ceil((non_cached_input * input_price + cached_input_tokens * cached_price + output_tokens * output_price) / 1,000,000)
cost = ceil(base * (1 - discount))

Balances are prepaid. Top up with an on-chain invoice from the billing page, or an admin can adjust the ledger. A request is rejected with 402 when the balance is at or below zero; an in-flight request can take the balance slightly negative.

Errors

Errors use the shape of the incoming API. Streaming requests return a normal JSON error before the stream commits; after commit, the error is delivered as an SSE event and the request is logged as failed.

Privacy

The proxy logs only usage metadata: timestamp, user, key id, model, provider, API dialect, stream flag, success, status, token counts, cost, latency, TTFT, tokens/sec, error code, and attempt count.

It does not store request bodies, response bodies, prompts, completions, headers, IP addresses, or user agents. Upstream credentials are encrypted at rest. API keys are stored as SHA-256 hashes and shown once.

Detailed request metadata is capped by row count so the database stays bounded at high throughput. Aggregate hourly and daily stats, cost, token, latency, error, and uptime totals are kept permanently.