Made in Nepal · Hosted API · v1.1

Jaynepal 1.1 is a Nepali language model you can call over HTTP.

Puja Set Nepal builds and hosts Jaynepal 1.1 — a Nepali-first model made in Nepal. It answers in Nepali, Romanised Nepali or English, and it is served from https://pujasetnepal.com/v1, so you can integrate it without renting or operating a GPU.

Read the documentation API reference

Model: Jaynepal 1.1
Languages: Nepali · Romanised · English
Interface: OpenAI-compatible · streaming
Serving: pujasetnepal.com

The model

Built for the language it speaks.

Jaynepal 1.1 was trained in Nepal for Nepali speakers. Three things follow from that, and they are the reasons to use it rather than a general model with Nepal bolted on.

Nepali first

Built for Nepali rather than translated into it. It answers in the language you wrote in — Devanagari, Romanised Nepali or English — and does not drift back to English unprompted.

Hosted for you

Running a 9B model needs a GPU, so we run it. You get an endpoint and an API key arrangement instead of a driver problem.

An interface you already have

The wire is chat/completions, so an OpenAI client library, a curl call or an agent framework works unchanged.

How it is served

An endpoint, not a deployment

You do not download weights, install CUDA or book a GPU by the hour. Your request goes to a single base URL; the model runs on our serving stack behind it and streams the answer back.

One base URL, one API shape — the same call works from a script, a backend or an agent.
The model id is all you pass to select it; there is nothing version-specific to wire up on your side.
Streaming is the default, so a long Nepali answer arrives as it is generated rather than in one late block.
Parameters and errors are documented, so failures are diagnosable instead of mysterious.

your app · script · agent
  │  POST https://pujasetnepal.com/v1/chat/completions
  │  { model: "jaynepal-1.1", messages: [...] }
  ▼
Puja Set Nepal  ·  hosted API        pujasetnepal.com
  │  OpenAI-compatible · SSE streaming
  ▼
Jaynepal 1.1  (9B)
  │  Nepali · Romanised Nepali · English
  ▼
streamed reply — delta frames, then done

One request, one model, one streamed response. Everything between your code and the answer is ours to keep running.

Capabilities

What you get when you call it.

Devanagari, properly

Nepali text is decoded and returned as characters, not escapes. Multi-byte boundaries are handled at the byte level, so replies stream without breaking words apart.

Streaming responses

Server-sent events from the first token. Long answers arrive progressively, which is what makes a Nepali reply at generation speed feel usable.

Markdown and code

Structure comes back as Markdown — headings, lists, tables and fenced code — so a client can render it without a translation layer.

Drop-in OpenAI client

Point any OpenAI SDK at the base URL and name the model. There is no custom protocol, no proprietary SDK and nothing to install.

Documented behaviour

Parameters, defaults, stream format and error codes are written down rather than discovered by experiment.

A hosted, not borrowed, endpoint

The model runs on our own serving stack behind a single base URL, so your integration does not depend on a notebook session.

Model card

The facts, in one table.

Model identity, size, languages and serving interface — the things you need before deciding whether to build on it. The full model card also lists what it is not good at.

Model: Jaynepal 1.1
Model id: jaynepal-1.1
Version: 1.1
Parameters: 9B
Languages: Nepali (Devanagari) · Romanised Nepali · English
Context: 8,192 tokens
Interface: OpenAI-compatible HTTP API · SSE streaming
Endpoint: https://pujasetnepal.com/v1
Trained by: Samir Puri, Bharatpur, Chitwan, Nepal

Quickstart

One request, no SDK.

Ask in Nepali and get Nepali back. Streaming is on by default; remove the final flag for a single JSON response.

curl -N https://your-endpoint/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $JAYNEPAL_API_KEY" \
  -d '{
        "model": "jaynepal-1.1",
        "messages": [
          { "role": "user", "content": "लमजुङ किन प्रसिद्ध छ?" }
        ],
        "stream": true
      }'

data: {"choices":[{"delta":{"content":"लमजुङ"}}]}
data: {"choices":[{"delta":{"content":" जिल्ला"}}]}
data: [DONE]

Use the base URL you were given for access — see access and status. The canonical production base URL is https://pujasetnepal.com/v1.

Full quickstart, parameters and streaming format

Status

Described accurately, not flatteringly.

A page that overstates what works is the fastest way to lose the developers it is written for. Here is the real state of things.

Working
Jaynepal 1.1 answers in Nepali, Romanised Nepali and English, over a streaming OpenAI-compatible API with documented parameters and errors.
Demo-grade
The public "try it" endpoint runs on a session-bound GPU host, so it goes away when the session ends and its address changes when it comes back. The documented production base URL is https://pujasetnepal.com/v1.
Planned
Self-serve API keys, per-key quotas, a published Nepali evaluation against comparable models, and a permanently hosted endpoint.

Start with the spec

Everything needed to integrate Jaynepal 1.1.

The model card, the request and response shapes, the streaming format, every parameter and every error code.

Read the documentation This website's source