// agent builder

Build an AI agent without writing code.

Describe the job, hand it your documents, pick the models. Fifteen minutes later you have an agent that remembers, reads and actually does the work — not another chat window.

How to build an AI agent, step by step

1. 01

   Describe the job in plain language

   No prompt engineering course required. Say what the agent should do, who it is for, and how it should sound. "Read our invoices and tell me who has not paid" is a valid specification.

2. 02

   Give it something to read

   Upload PDFs, spreadsheets, notes or past papers. The agent grounds its answers in those documents instead of guessing, and it handles Devanagari as well as English.

3. 03

   Pick which models it may use

   GPT-5, Claude, Gemini, DeepSeek and 300+ more. An agent can switch per task — a cheap model for extraction, a strong one for judgement — instead of paying top rates for everything.

4. 04

   Connect tools over MCP (optional)

   Model Context Protocol lets the agent reach systems you already run: a database, an internal API, a file store. This is the step that turns an explainer into something that does the work.

5. 05

   Run it, and pay only while it runs

   No server to rent, nothing to deploy. Credits are spent while the agent is working and not otherwise, so an agent used twice a week costs roughly nothing.

Things people actually build

• Chat with a folder of PDFs

  Point it at contracts, manuals or research and ask questions across all of them at once.

  Read more →

• A study agent that remembers

  It tracks which topics you keep getting wrong and drills those, session after session.

  Read more →

• Customer support triage

  Reads incoming messages, drafts a reply in your tone, and flags anything it should not answer alone.

• A research assistant

  Searches, reads, and returns a summary with the sources rather than a confident paragraph with none.

• Invoice and admin chasing

  Extracts totals and dates from documents, then drafts the follow-up messages.

  Read more →

• Something nobody has built yet

  The builder is general. If you can describe the job and supply the documents, it can usually be assembled.

  Read more →

Memory, RAG and MCP — what each one is for

Three capabilities do most of the work, and they solve different problems. Long-term memory is what stops you re-explaining your context every session. Document RAG grounds answers in files you supply, which is what makes an agent trustworthy on your own material rather than plausible-sounding in general. MCP tools let it reach systems you already run.

Most agents need the first two and not the third. Longer explanations in memory and RAG explained and what MCP actually is.

What it costs

No subscription. Start free at Rs 0, top up from Rs 100 up to Rs 3,500, and spend credits across any model. The agent consumes credits while it runs and not otherwise, so an agent you use twice a week does not cost the same as one running continuously. Pay via eSewa, Khalti or Fonepay — no dollar card, no VPN. Details on the pricing page.

Frequently asked questions

Do I need to know how to code to build an AI agent?

No. The builder is no-code: you describe the task in plain language, upload any documents the agent should know, and choose which models it may use. Developers can go further by connecting custom MCP servers, but that is optional and not required to get a working agent.

What is the difference between an AI agent and a chatbot?

A chatbot answers the message in front of it and forgets you afterwards. An agent keeps long-term memory across sessions, reads documents you have given it, and can call tools to actually perform a task rather than describing how you would do it yourself.

What does it cost to run an AI agent?

There is no subscription. You start free at Rs 0, then top up credits from Rs 100, and the agent spends credits only while it is running. Rs 1 buys 10 credits, larger top-ups include bonus credits, and credits never expire while your account is active.

What is MCP and do I need it?

MCP (Model Context Protocol) is an open standard for connecting an AI to external tools and data sources. You do not need it for most agents. You want it when the agent has to reach a system you already run, such as an internal database or API, rather than only the documents you uploaded.

Can an agent read Nepali documents?

Yes. Agents handle Devanagari text, so documents such as government notices, legal text or Nepali study material can be uploaded and queried directly, with answers in either language.