Compare

Comparisons

Build your own AI agent platform, or use promptev?

By Faisal SaeedUpdated 2 min read

In short

Building your own AI agent platform is achievable, and it fits your company exactly. The demonstration takes about two weeks. The work that follows is the product: connectors and their token refreshes, retrieval that respects each person's permissions, re-indexing when files change, scanned documents, approvals, audit, channels, billing and an app other people can use, then maintaining all of it. Build if the agent platform is your product. Use promptev if it is not. Its context engine is open source, so you can still read it, run it and keep it.

Almost every engineering team can build the first version. The question is what you want your engineers to maintain in two years.

Where building is the better choice

  • The agent platform is your product. Then the control is worth the cost.
  • A very unusual requirement. Something no product can be set up to do.
  • You already have the team. Engineers who will own connectors, retrieval, security and the user interface for the long term.

What the build really includes

The model call is the small part. The list after the demonstration looks like this:

  1. Connectors. Sign-in for each source, token refresh, rate limits, and changes each time a provider changes its API.
  2. Sync. Re-read what changed, remove what was deleted, and follow moved files.
  3. Permissions. Carry who may see what into every search, and keep it correct when sharing changes.
  4. Reading documents. Scanned PDFs, spreadsheets, slides and images.
  5. Approvals. Rules for each action, a queue, a way to decide by e-mail or in chat, and expiry.
  6. Audit. Every run, approval and settings change, and who did it.
  7. Channels. A chat app, Slack, WhatsApp, a website widget, an API.
  8. Billing and limits. Who used what, and a stop before a balance runs out.
  9. An app that other people can use. Roles, projects, sign-in with your identity provider.

Where promptev differs

  • All of that is included and maintained. You set it up; you do not build it.
  • The part you would most want to own is open. The context engine is open source under Apache-2.0.
  • You keep control of the data. A workspace can use its own database, its own model keys and its own storage, and an Enterprise licence can run in your own cloud.
  • You can start today. promptev has no seats and no plans. A new workspace gets 2,500 free credits with no card. After that a credit costs $0.005, and you buy from $5.

Which to choose

Choose to build if this is your product, or you have a requirement that no platform can meet.

Choose promptev if the agent is a tool for your company. Your engineers then spend their time on what makes your company different.

Frequently asked questions

How long does it take to build an AI agent on company data?

A working demonstration takes about two weeks. A product that many people can use safely takes much longer, because most of the work is connectors, permissions, approvals, audit and maintenance, not the model call.

What is the hardest part to build yourself?

Permissions. Each answer must use only the documents the asker may open, and that must stay true when sharing changes, when a person leaves, and across search methods. It is easy to get right in a demo and hard to keep right.

Can we build on promptev's engine instead of from zero?

Yes. The context engine is open source under Apache-2.0, with Python and TypeScript libraries. It gives you hybrid and graph retrieval with access rules on every passage, and you build your own product around it.

What does promptev cost compared with building?

promptev has no seats: 2,500 free credits to start, then $0.005 a credit for the work agents do. Building costs engineers' time first and maintenance for as long as the system runs.