// Docs · Migrate · qua- → sl-

qua- → sl-.

If you are already calling Quatarly through the OpenAI SDK, this is a four-minute migration. The body shape, the SDK, the exception surface — all unchanged. Three things change: base URL, key prefix, error envelope.

// What you are not doing You are not rewriting prompts. You are not changing model IDs (most map 1:1). You are not changing your retry logic. You are not switching SDKs.

01The three changes, side-by-side

What	Quatarly	SLASHED
Base URL	https://api.quatarly.ai/v1	https://api.slashed.pro/v1
Auth header	X-Quatarly-Token: qua-...	Authorization: Bearer sl-...
Key prefix	qua-...	sl-...
Error envelope	{"err_code", "err_msg"}	{"error":{"message","type","code","param"}}
SDK	Official OpenAI SDK	Official OpenAI SDK (unchanged)
Request body	OpenAI v1 spec	OpenAI v1 spec (unchanged)
Response body	OpenAI v1 spec	OpenAI v1 spec (unchanged)

02The five-step migration

Step 01

Get a SLASHED key.

Sign in at dashboard.slashed.pro. Create a key under Keys → New. Copy the sl-... secret once — it is shown a single time. Optional but recommended: name it migration-from-quatarly and scope it to the model IDs you intend to use.

Step 02

Swap the env var.

If you have a single env var (e.g. OPENAI_API_KEY, QUATARLY_API_KEY, LLM_KEY) that holds the bearer token, set it to the new sl- value. If you have two env vars (key + base URL), set both.

# .env (or your deploy secrets manager)
OPENAI_BASE_URL=https://api.quatarly.ai/v1
OPENAI_API_KEY=qua-1234abcd_secretsecret...

OPENAI_BASE_URL=https://api.slashed.pro/v1
OPENAI_API_KEY=sl-a7f3c2b4_secretsecret...

Step 03

Update the SDK constructor.

If your code passes base_url / baseURL in source code (rather than via env), update both. The SDK choice does not change — keep the official openai package.

# Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.quatarly.ai/v1",
    api_key="qua-...",
    base_url="https://api.slashed.pro/v1",
    api_key="sl-...",
)

// Node
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.quatarly.ai/v1",
  apiKey: "qua-...",
  baseURL: "https://api.slashed.pro/v1",
  apiKey: "sl-...",
});

Step 04

Drop any custom auth-header logic.

If you have code that injects X-Quatarly-Token via a custom HTTP middleware (because the OpenAI SDK does not set that header natively), remove it. SLASHED uses the standard Authorization: Bearer header that the SDK sets for you automatically.

# BEFORE — Quatarly required a custom header injection
client.default_headers["X-Quatarly-Token"] = quatarly_key

# AFTER — nothing. The SDK handles Authorization: Bearer for you.

Step 05

Update any error-parsing code.

If you parse the raw HTTP error body (rather than relying on the SDK's APIError exception class), switch from Quatarly's {"err_code", "err_msg"} envelope to OpenAI's {"error": {"message", "type", "code", "param"}}.

# BEFORE
try:
    resp = client.chat.completions.create(...)
except Exception as e:
    body = json.loads(e.response.text)
    log.error(body["err_code"], body["err_msg"])

# AFTER
try:
    resp = client.chat.completions.create(...)
except Exception as e:
    body = json.loads(e.response.text)
    log.error(body["error"]["code"], body["error"]["message"])

If you only catch the SDK's openai.APIError / RateLimitError / AuthenticationError classes, no change is needed. The SDK maps both envelopes onto the same exception types.

03Model ID mapping

Most Quatarly model IDs map 1:1 to SLASHED. The exceptions are listed below — if your code references a Quatarly-only ID, swap to the closest SLASHED equivalent.

Quatarly ID	Closest SLASHED ID	Note
gpt-5.4-mini	gpt-5.4-mini	1:1
gpt-5.4	gpt-5.4	1:1
gpt-5.5	gpt-5.5	1:1
claude-opus-4.7	claude-opus-4-7	Dot → dash in segment
claude-sonnet-4.6	claude-sonnet-4-6	Dot → dash in segment
gemini-3.5-flash	gemini-3.5-flash	1:1
gemini-3.1-pro	gemini-3.1-pro-preview	SLASHED exposes the preview-tagged ID
gemini-2.5-pro	gemini-2.5-pro	1:1

For the full set of valid SLASHED IDs, hit GET /v1/models or see /docs/models.

04Smoke test

Confirm the swap took with a one-line cURL. If you get a 200 with a model reply, you are done.

curl https://api.slashed.pro/v1/chat/completions \
  -H "Authorization: Bearer $SLASHED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-mini","messages":[{"role":"user","content":"reply: migrated"}]}'

05Run both side-by-side (optional)

If you want a few days of dual-write before fully cutting over, the cleanest pattern is two SDK clients in the same process. The OpenAI SDK is stateless — instantiating two clients with different base_url / api_key is supported.

from openai import OpenAI

quatarly = OpenAI(
    base_url="https://api.quatarly.ai/v1",
    api_key=os.environ["QUATARLY_KEY"],
)
slashed = OpenAI(
    base_url="https://api.slashed.pro/v1",
    api_key=os.environ["SLASHED_KEY"],
)

# Send the same prompt to both. Compare outputs & p99 latency.
# When confident, drop the quatarly client and promote slashed.

06Things that will not change

Your prompts. Same model, same context, same output distribution.
Your retry logic. The SDK's default retry on 429/5xx works on both. See /docs/errors §05.
Your streaming code. stream=True on both, SSE shape is OpenAI-spec on both.
Your tool-use code. Same array shape, same tool-call response surface.
Your token-counting code, if you use tiktoken for OpenAI families.

// One known difference at migration Quatarly's older accounts surface a X-Quatarly-Request-Id header on every response. SLASHED does not document a request-id header today. If your observability layer reads that header, switch to relying on the response body's id field (e.g. chatcmpl-...), which both gateways emit.

07If you get stuck

Auth error? → See /docs/authentication and /docs/errors §03.
Unknown-model error? → See /docs/models and the mapping table above (§03).
Need the dashboard? → See dashboard.slashed.pro for live usage, key management, and per-seat caps.
Anything weird? → The dashboard ledger shows every request that has reached the gateway, with status, model, and token count.