---
title: "About Tarka — AI and agent infrastructure from Nepal"
description: "Tarka is an AI and agent infrastructure company based in Kathmandu, building model APIs and runtimes for developers and organizations."
url: "https://tarkahq.com/about/"
---

About Tarka

# An AI and agent infrastructure company based in Kathmandu.

Tarka gives developers one place to run models, agents, document and speech tools, batch jobs, and code sandboxes.

We are building the infrastructure we wanted to use ourselves: familiar APIs, clear usage, and enough control to move a serious workload into production.

[Explore Tarka](https://tarkahq.com/inference/) [Talk to us](https://tarkahq.com/enterprise/#enterprise-intake)

Kathmandu, Nepal

OpenAI-compatible APIs

Models and compute runtimes

Self-serve and enterprise

The company

## We make AI infrastructure easier to use and easier to account for.

Using a model is straightforward. Operating the surrounding system usually is not. Teams end up managing several providers, different credentials, separate billing, and a growing collection of deployment scripts.

Tarka brings that work together. A team can call different models through one interface, run longer jobs on managed compute, give an agent a controlled runtime, and see requests and spend in the same account.

We do not require customers to pretend every workload is the same. A prototype can use a shared endpoint. A larger or more sensitive system can move to a private route, dedicated capacity, or approved customer infrastructure.

What we build

## One account across models, compute, and data.

Inference, utilities, managed compute, and storage share the same account and usage view.

- **Model inference**: Open and provider models through an OpenAI-compatible endpoint, with usage and model identity kept visible.
- **Model inference**: [Browse models →](https://tarkahq.com/inference/)
- **Documents and speech**: OCR, extraction, transcription, and text to speech through the same account.
- **Documents and speech**: [Browse utilities →](https://tarkahq.com/inference/#speech-and-documents)
- **Agents and batch jobs**: Managed runtimes for tool-using agents, evaluations, document pipelines, scheduled tasks, and other long-running work.
- **Agents and batch jobs**: [Explore agents →](https://tarkahq.com/compute/#hosted-agents)
- **Sandboxes and storage**: Controlled code execution with the files, repositories, datasets, and artifacts a workload needs.
- **Sandboxes and storage**: [Explore sandboxes →](https://tarkahq.com/compute/#sandboxes)

Why Nepal

## Building a global platform from Nepal

Tarka is built in Nepal. That means working near developers and institutions whose languages, procurement constraints, and infrastructure needs are often treated as edge cases by global platforms.

It also means building for the wider market from the start. The API should be familiar to a developer anywhere; the deployment options should be specific enough for the customer in front of us.

Operating principles

## What customers should expect from Tarka

Three practical rules guide the platform and our work with customers.

1. 01

   ### Keep the interface familiar.

   A customer should not need to rewrite an application to try a different model or deployment.
2. 02

   ### Show what is actually running.

   Model, provider, route, usage, and cost should remain visible rather than disappear behind a generic AI label.
3. 03

   ### Do not promise a boundary we cannot operate.

   Location, retention, support, capacity, and security commitments belong in the design and the agreement.

## Start with the API or bring us a workload.

Developers can create an account and begin with the public platform. Teams with private models, data constraints, or dedicated capacity needs can talk directly with us.

[Create an account](https://tarkahq.com/console?auth=signup) [Private deployments](https://tarkahq.com/enterprise/)