Models

The API speaks OpenAI's chat completions format at https://api.abhashchakraborty.tech/v1, so the OpenAI SDKs work with it. Every request needs only your key.

1. Get a key

Run api key in the terminal. It checks your browser, then shows the key once with the commands below. One key per person: a new key switches the old one off, and every key stops working 3 hours after it's made.

2. Keep it in your shell

Linux and macOS:

export ABHASH_API_KEY="<your key>"

Windows PowerShell:

$env:ABHASH_API_KEY = "<your key>"

3. Ask a model

curl https://api.abhashchakraborty.tech/v1/chat/completions \
  -H "Authorization: Bearer $ABHASH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"ophyx-350m-v2","messages":[{"role":"user","content":"Write is_palindrome(s)"}]}'

In Windows PowerShell, where curl is a different command:

$body = @{
  model = "ophyx-350m-v2"
  messages = @(@{ role = "user"; content = "Write is_palindrome(s)" })
} | ConvertTo-Json -Depth 4
$reply = Invoke-RestMethod -Method Post -Uri "https://api.abhashchakraborty.tech/v1/chat/completions" `
  -Headers @{ Authorization = "Bearer $env:ABHASH_API_KEY" } `
  -ContentType "application/json" -Body $body
$reply.choices[0].message.content

Stream the answer as it's written:

curl -N https://api.abhashchakraborty.tech/v1/chat/completions \
  -H "Authorization: Bearer $ABHASH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"ophyx-350m-v2","stream":true,"messages":[{"role":"user","content":"Write is_palindrome(s)"}]}'

With the OpenAI SDK for Python (pip install openai):

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.abhashchakraborty.tech/v1", api_key=os.environ["ABHASH_API_KEY"])
reply = client.chat.completions.create(
    model="ophyx-350m-v2",
    messages=[{"role": "user", "content": "Write is_palindrome(s)"}],
)
print(reply.choices[0].message.content)

From JavaScript or TypeScript, with no SDK:

const res = await fetch("https://api.abhashchakraborty.tech/v1/chat/completions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.ABHASH_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "ophyx-350m-v2",
    messages: [{ role: "user", content: "Write is_palindrome(s)" }],
  }),
});
const data = await res.json();
console.log(data.choices[0].message.content);

The models this API serves right now:

curl https://api.abhashchakraborty.tech/v1/models

Request body

• model: required, an id from /v1/models
• messages: the whole conversation: earlier user and assistant turns reach the model, so follow-ups work, and it keeps the newest that fit its context. System messages are ignored. The last user message can be up to 2,000 characters
• max_tokens: up to 768, the default
• temperature: 0 to 1.5, default 0.2
• stream: true for server-sent events; add "stream_options":{"include_usage":true} for token counts at the end
• concise: true by default: plain code with few comments. Send false for docstrings and comments

Limits

• One request at a time per key; 5 a minute, 50 an hour and 100 a day.
• Each day a key also shares its owner's 60 messages with the terminal chat, and the whole site has a daily cap, after which the API rests until midnight UTC.

Errors

Errors use OpenAI's shape, { error: { message, type, code } }, and the message says what to do.

• 401 invalid_api_key · key_expired: run api key for a new one
• 404 model_not_found: check the id against /v1/models
• 429 concurrent_requests · rate_limit_exceeded: wait for the time in Retry-After
• 503 api_off · model_offline: the API or that model is switched off