Setting up Cisely

Cisely is a browser app with an agent in it. Sign in and you are set up — there is nothing to install, and the fastest path to something real is a conversation.

That agent is Paroksh. You describe your business; they draft the model it is built from — the mission and beliefs, the personas, the domain, the operations — and hand each piece back for you to correct. You keep the judgement; they do the writing down.

Everything below is optional. The command-line tool and the agent connector exist for people who want a terminal or their own AI agent, and they are worth having, but nothing needs them on day one.

flowchart LR
  you["you"]
  app["Cisely<br/>jharokha.cisely.ai"]
  p(["Paroksh<br/>the agent inside"])
  api["the platform"]
  cli["cisely CLI<br/>(optional)"]
  agent["your own AI agent<br/>(optional)"]
  mcp{{"agents.cisely.ai/mcp"}}
  you --> app --> p --> api
  you -.-> cli -.-> api
  you -.-> agent -.-> mcp -.-> api

Start here

  1. Create your Cisely account — an email and a one-time code.
  2. Working with Paroksh — what the agent does, and where they stop.
  3. The Workshop — the eight views of the application you are building.

That is the whole setup. Read on only if you want the extras.

The optional tools

ToolWhat it addsWho it is for
cisely CLIDrive your project from a terminal: create, author, validate, build, pull generated code.People who would rather type than click, and anyone scripting Cisely into their own pipeline.
MCP connectorLets an AI agent you already use — Claude, Codex — read and write your Cisely model on your behalf.People with an agent and a workflow they like. Paroksh does this inside Cisely already.

What you need

  • A browser. That is the whole requirement for the main path.
  • An email address you can receive a code at. There are no passwords.
  • (Optional) macOS or Linux for the CLI, or WSL on Windows.

You do not need a language runtime, a framework, a database, or a build toolchain to start. Code generation happens on Cisely's servers, so there is nothing to install locally just to see your model turn into a running system.

One thing worth knowing before you begin

Paroksh writes drafts. Nothing they produce counts until you activate it.

That is not a safety disclaimer bolted on the side — it is the division of labour the whole product is built around. They are fast and they are frequently wrong about your business, because they have never met your customers. You are slow and you are right. The loop only works if their output arrives as a proposal you correct, which is why it does.

The quickest way to understand them is to spend ten minutes describing your business and read what comes back. The parts they get wrong are the parts only you could have known — and seeing exactly which parts those are is the fastest orientation available.

Once you are set up, head to Introduction to Vishwakarma to learn the language your model is written in — the one Paroksh writes, and the one you will read and correct.