Madhuram API

Quickstart

Create an API key from the chat interface, sign in, open Settings → API, and select Create API key. The key is shown once and cannot be retrieved afterwards, so store it somewhere safe.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_KEY",
    base_url="https://api.maruthlabs.com/v1"
)

response = client.chat.completions.create(
    model="madhuram",
    messages=[{"role": "user", "content": "Explain gradient descent in two sentences."}]
)

print(response.choices[0].message.content)

curl https://api.maruthlabs.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "madhuram",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_KEY",
  baseURL: "https://api.maruthlabs.com/v1"
});

const r = await client.chat.completions.create({
  model: "madhuram",
  messages: [{ role: "user", content: "Hello" }]
});

console.log(r.choices[0].message.content);

Authentication

Every request requires a bearer token in the Authorization header:

Authorization: Bearer sk-mrth-...

Keys are issued per account, one active key at a time. Regenerating a key revokes the previous one immediately. Only a hash of your key is stored on our side, which means we cannot recover it for you, if it is lost, generate a new one.

Endpoints

Base URL: https://api.maruthlabs.com/v1

POST/v1/chat/completions

Generates a completion for a conversation. Accepts and returns the standard OpenAI chat-completions shape, with or without streaming.

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1785312496,
  "model": "madhuram",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "..." },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 57,
    "completion_tokens": 10,
    "total_tokens": 67
  }
}

GET/v1/models

Lists available models.

{
  "object": "list",
  "data": [{ "id": "madhuram", "object": "model", "owned_by": "maruthlabs" }]
}

GET/v1/usage

Returns your consumption against the current day's quota.

{
  "tier": "free",
  "day": "2026-07-29",
  "requests_used": 12,
  "requests_limit": 1000,
  "tokens_used": 4820,
  "tokens_limit": 200000,
  "resets": "00:00 UTC"
}

Parameters

Parameter	Type	Description
messages	array	Required. Conversation history. Each entry has a role (system, user, or assistant) and content.
model	string	Use "madhuram". Other values are ignored.
max_tokens	integer	Maximum tokens to generate. Capped at 4096. Values of -1 or 0 are treated as "use the maximum".
temperature	number	Sampling temperature. Lower is more deterministic; higher is more varied.
top_p	number	Nucleus sampling threshold.
top_k	integer	Restricts sampling to the k most likely tokens.
repetition_penalty	number	Multiplicative penalty on repeated tokens. 1.0 is neutral; useful range is roughly 1.0 to 1.3.
presence_penalty frequency_penalty	number	Additive OpenAI-style penalties. Do not stack these with repetition_penalty or output degrades.
stop	string / array	Sequences that halt generation. A bare string is accepted and normalised.
seed	integer	Seeds sampling for reproducible output.
stream	boolean	Streams tokens as server-sent events. Defaults to false.

Streaming

Set stream: true to receive tokens as they are generated, in the standard OpenAI server-sent-event format terminated by data: [DONE].

stream = client.chat.completions.create(
    model="madhuram",
    messages=[{"role": "user", "content": "Write a haiku about servers."}],
    stream=True
)

for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

Rate limits

Free-tier limits, applied per account:

Limit	Value	Window
Requests per minute	60	Rolling
Requests per day	1,000	Resets 00:00 UTC
Tokens per day	200,000	Resets 00:00 UTC
Max tokens per request	4,096	Per request
Context window	16,384	Prompt + completion

Responses carry X-RateLimit-Remaining-Requests and X-RateLimit-Remaining-Tokens headers so clients can self-throttle. Building something that needs more? Get in touch, higher limits are available on request.

Errors

Status	Meaning	What to do
400	Malformed request	Check that messages is present and well-formed, and that the prompt fits the context window.
401	Missing, invalid, or revoked key	Verify the Authorization header. If the key was regenerated, update it.
429	Rate limit or daily quota exceeded	Back off and retry. The Retry-After header gives the wait in seconds.
503	Inference backend unavailable	Transient. Retry with exponential backoff.

Data & privacy

Requests and responses are recorded for usage metering, debugging, and abuse investigation. They are used to train future models only if you opt in, the setting is off by default and can be changed at any time under Settings → API in the chat interface.

The API is free for Madhuram-v0.6. We will give notice before introducing pricing.