Documentation

Harvanera gives you one API key and one balance for every model in the catalog. The API is OpenAI-compatible: any SDK or tool that works with OpenAI works with Harvanera — change the base URL and the key.

Quick start

Create an API key in your dashboard. New accounts get a welcome credit to try things out.
Pick a model in the catalog and copy its ID, for example deepseek/deepseek-v3.1.
Send a request:

curl https://harvanera.com/v1/chat/completions \
  -H "Authorization: Bearer sk-harvanera-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v3.1",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

The same with the official OpenAI SDK:

from openai import OpenAI

client = OpenAI(base_url="https://harvanera.com/v1", api_key="sk-harvanera-YOUR_KEY")
response = client.chat.completions.create(
    model="deepseek/deepseek-v3.1",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://harvanera.com/v1", apiKey: "sk-harvanera-YOUR_KEY" });
const response = await client.chat.completions.create({
  model: "deepseek/deepseek-v3.1",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);

Prefer to try first? Use the chat in your dashboard — it runs on the same balance and shows the cost of every reply.

Models and keys

You don't buy models separately. One key works with every available model: the model is chosen per request with the model field, and you pay only for the tokens of the model you called.

The list of models with prices: GET https://harvanera.com/v1/models (no key needed) or the catalog.
Models marked temporarily unavailable return 503 model_unavailable. Their IDs are reserved, so your code will start working without changes once they are connected.
A key can be restricted to specific models, capped with a monthly spend limit, or paused — see API keys.

Streaming

Set "stream": true to receive the reply as Server-Sent Events, exactly as with OpenAI. To get token usage in the last chunk, add "stream_options": {"include_usage": true}.

stream = client.chat.completions.create(
    model="qwen/qwen3-coder",
    messages=[{"role": "user", "content": "Write a haiku about APIs"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

If you stop reading a stream midway, you are charged only for the tokens generated before the disconnect.

Pricing and balance

Prices are per 1M input and output tokens and are listed on each model's page. Your balance is kept in USD and shown in the currency you choose.
Every response has an x-request-id header; non-streaming responses also have x-cost-usd with the exact charge.
Check the balance from code: GET https://harvanera.com/v1/balance with your key.
Before each request we check that the balance covers the worst case (the prompt plus max_tokens of output). If it doesn't, the API returns 402 insufficient_balance. Setting max_tokens lowers that estimate.
Every request, with its model, tokens and cost, is listed on the Activity page and can be exported to CSV.

Errors

Errors use the OpenAI format: {"error": {"message": "...", "type": "...", "code": "..."}}.

Status	Code	What to do
400	`invalid_request`	Fix the request body: it must be JSON with a non-empty `messages` array.
401	`invalid_api_key`	The key is wrong, revoked or paused.
402	`insufficient_balance`	Top up the balance or lower `max_tokens`.
402	`key_limit_exceeded`	The key reached its monthly limit — raise it on the API keys page.
403	`model_not_allowed`	The key is restricted to other models.
404	`model_not_found`	Check the model ID against `GET /v1/models`.
429	`rate_limited`	Too many requests — retry after the `Retry-After` header.
429	`key_paused`	The key is paused after many failed requests in a row; wait and fix the requests.
502	`upstream_error`	The model provider failed. Retry — we already tried the backup providers.
503	`model_unavailable`	The model is temporarily unavailable.

Limits

60 requests per minute per key by default.
After 20 failed requests (4xx) within a minute the key is paused: for 1 minute, then 10 minutes, 1 hour and 12 hours if it keeps happening. This protects your balance from runaway scripts.

Integrations

Anything that supports an OpenAI-compatible provider works. You always need three things: the base URL https://harvanera.com/v1, your key and the model ID.

Cursor — Settings → Models: enter the key in OpenAI API Key, enable Override OpenAI Base URL and set it to the base URL, then add the model ID.
Cline / Roo Code — API Provider: OpenAI Compatible; Base URL, API Key and Model ID as above.
n8n — create OpenAI credentials and set Base URL to the base URL; pick the model ID in the node.
LangChain — ChatOpenAI(base_url="https://harvanera.com/v1", api_key="sk-harvanera-…", model="deepseek/deepseek-v3.1").

Menu names in third-party tools change between versions; if something doesn't match, look for "OpenAI-compatible" or "custom base URL" in the tool's settings.