Htvag Agent — frequently asked questions

Htvag Agent is an autonomous AI coding agent: a desktop application that reads your repository, edits real files, runs your build and test commands, reads the output, and keeps iterating until the outcome you described works. It runs on Windows 10/11 and Linux, and there is a browser chat as well.

Htvag, a company based in India.

No. It is a separate desktop app that operates on a folder, so it works alongside VS Code, JetBrains, Neovim or anything else, on any repository. This is the main practical difference from Cursor, which is a fork of VS Code you adopt wholesale.

Web application work: Next.js, React, TypeScript, HTTP APIs, authentication and databases. It understands the server/client component boundary and where a secret must never end up.

The free plan is ₹0. Paid plans are ₹399 (Starter), ₹799 (Pro) and ₹1299 (Max) per month. Every plan is a flat monthly price with no overage billing.

Yes, there is a permanently free plan and it is the full agent — file edits, shell access, test runs. It gives 5 credits per day, up to 20 by watching ads in the app, and requires no credit card.

One credit is roughly 10,000 tokens of work. A small fix costs a fraction of one; a feature spanning several files with a build-and-fix loop costs a few. Credits reset every day at 00:00 UTC and do not roll over.

The agent pauses until the next daily reset. Nothing bills automatically and there is no overage charge — you can upgrade at any time if you want more room.

Checkout runs through Razorpay and accepts UPI, RuPay, domestic and international cards, net banking and wallets. Prices are set in rupees, so there is no currency conversion or forex markup.

Yes. Upgrades apply immediately; downgrades and cancellations take effect at the end of the period you have already paid for.

At the tier where agent work is actually usable, yes. ₹399/month entry and ₹1299/month flat at the top, against roughly ₹3,237 for GitHub Copilot Pro+, ₹4,980 for Cursor Pro+, ₹8,300+ for Claude Max, and up to ₹16,600 for heavy ChatGPT use.

No. The agent reads and edits files locally and runs commands in a shell on your machine. Only the instruction and the specific context needed to answer it are sent for processing, and none of it is stored. Whole repositories are never uploaded.

No. It is confined to the folder you open for the whole session and cannot read or write anything outside it.

No. Nothing sent for processing is retained or used for training.

Yes. Reasoning, tool calls and diffs stream live as the work happens, so a wrong plan is visible at step two rather than after the fact. You can stop a run at any point.

No. Every account, free included, comes with a daily allowance through the built-in gateway. There is no key to obtain or paste and no model to select.

Windows 10 and 11, and Linux via AppImage. macOS is not shipped yet. The web chat works in any browser, and there's a CLI too.

The desktop app opens your browser at this site, you approve the request, and the app is signed in automatically. There is nothing to copy or paste.

File reads, edits and your builds are local, but the model itself is remote, so an internet connection is required to give it instructions.

Copilot is an extension inside your editor and meters premium and agent-mode requests. Htvag Agent is a standalone agent with a flat daily allowance that runs your build and keeps fixing until it passes, at about 2.5× less than Copilot Pro+.

Cursor asks you to switch editors; Htvag Agent runs beside the one you have. Cursor's in-editor experience and tab completion are better; Htvag Agent is about 3.8× cheaper at the agent-capable tier and has no usage cap mid-session.

Same category — both are autonomous agents. Claude Code reasons better on the hardest refactors and has a larger ecosystem; Htvag Agent costs about 6.4× less, ships both a desktop UI and a terminal CLI, and bills in rupees with a free tier.

A chat window cannot open your repository, apply its own suggestion, or run the test that would prove the suggestion wrong. An agent does all three, which is why it finishes tasks rather than producing drafts.