Getting started
Chat with models right on the website, or plug your Relé key into your own code and coding tools. One balance covers everything.
- Create an account. Your first API key is created automatically — copy it on the keys page.
- Top up your balance and pick a model in the catalog.
- The Relé address is already in the examples below. Replace only the model ID and the API key.
API
Relé speaks the OpenAI Chat Completions format, so any OpenAI SDK or HTTP client works. Use your Relé key and a model ID from the catalog.
# macOS / Linux
export RELE_API_KEY="YOUR_API_KEY"
curl https://api.rele.click/v1/chat/completions \
-H "Authorization: Bearer $RELE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3",
"messages": [
{"role": "user", "content": "Write a Python function that removes duplicates while preserving order."}
],
"max_tokens": 1024,
"stream": false
}'The answer is in choices[0].message.content. To switch models, change the model field. For streaming, set stream to true and read Server-Sent Events until [DONE].
Open chatClaude Code
Point Claude Code at Relé with three environment variables. Set the address without /v1 — Claude Code adds /v1/messages itself. Use a Claude model ID from the catalog.
# macOS / Linux
export ANTHROPIC_BASE_URL="https://api.rele.click"
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
export ANTHROPIC_MODEL="claude-sonnet-5"
claude
# PowerShell
$env:ANTHROPIC_BASE_URL="https://api.rele.click"
$env:ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
$env:ANTHROPIC_MODEL="claude-sonnet-5"
claudeClaude Code gateway guide ↗Codex CLI
Add Relé as a provider in ~/.codex/config.toml on macOS/Linux or %USERPROFILE%\.codex\config.toml on Windows. This goes into your user config, not a project’s .codex folder.
model = "gpt-6.1-sol"
model_provider = "rele"
[model_providers.rele]
name = "Relé"
base_url = "https://api.rele.click/v1"
experimental_bearer_token = "YOUR_API_KEY"
wire_api = "responses"Paste your full Relé API key instead of YOUR_API_KEY, save the file and restart Codex. To use another model, change the first line; keep model_provider = "rele" as is.
Kimi Code
Kimi Code works with any OpenAI-compatible provider. In the desktop app open Settings → Model Providers and add Relé: OpenAI protocol, your Relé key, base URL https://api.rele.click/v1 and a model ID from the catalog. For the CLI, add the provider to ~/.kimi-code/config.toml:
default_model = "rele"
[providers.rele]
type = "openai"
base_url = "https://api.rele.click/v1"
api_key = "YOUR_API_KEY"
[models.rele]
provider = "rele"
model = "glm-5.3"
max_context_size = 128000Kimi Code provider guide If something goes wrong
401Check your key. It may be missing, revoked or copied incorrectly.
402Not enough balance. Part of it may be reserved for requests that are still running.
403This key can't make this request. Make sure the key belongs to your account; if the error repeats, write to support.
404Unknown model or wrong address. Copy the exact model ID from the catalog and check the base URL.
429Too many requests at once. Wait a moment and lower the number of parallel requests.
400Something in the request is off. If the context is too long, start a new session or shorten the input.
5xxThe model is temporarily unavailable. Try again in a minute.
Don't retry requests that run tools automatically — check the history first so actions and charges aren't duplicated.
Pay for what you use
No subscription. Top up your balance and each request is charged by the tokens it actually used. The catalog shows the price of every model in rubles per 1 million tokens.
The website chat has its own, lower tariff: the price per 1 million tokens is shown under the message box for the selected model. Chat requests with photos, including photos earlier in the conversation, cost 5% more.
View pricesRelé tokens
Your balance is shown in Relé tokens: 1 million Relé tokens = 15 ₽, so a 100 ₽ top-up gives about 6.67 million.
A request spends Relé tokens in proportion to the model's own tokens: input, cached input and output are charged at the same rate. For example, 200,000 tokens on a model priced at 60 ₽ per 1M cost 12 ₽.
Before a request starts, an estimated amount is reserved. When it finishes, only the actual usage is charged and the rest of the reserve is returned.
In API responses, usage.input_tokens, usage.output_tokens and usage.total_tokens are the model's real tokens, not Relé tokens.
Web search in chat
When you ask about recent events or explicitly ask to search, the website chat looks things up on the web and shows source links. Regular questions and writing tasks don't trigger a search.
Search results are added to the request, so it uses more input tokens. Some models search on the provider's side for an extra fee — it's shown next to the model in the chat. In the API, you connect your own tools — automatic search only works in the website chat.