OpenWOP openwop.dev

For companies

For companies.

You want AI agents doing real work for your company: checking supplier invoices, answering customers, getting a launch out the door. This page is for the person deciding where that work should live. It makes the case for starting from the OpenWOP reference app, under your own brand and on infrastructure you choose, and it is plain about what you take on when you do.

A 117-second narrated film, with English captions. A company wants AI agents doing real work, and has to decide where that work should live. Buying a platform puts the workflows inside someone else's product; building your own takes months on approvals, a record and undo. The film shows a third way: the OpenWOP reference app, which is open source, downloaded, started with one command and rebranded as the placeholder company Example Co. Real screens from that branded app show a workflow on the canvas, a payment waiting for finance to approve, the recorded event stream with a fork on every step, and ten example agents. Your brand, your cloud and your keys follow, with agents and workflows moving to another host while the keys stay behind. It ends plainly: you run it, so keeping it running is yours too, with no service contract behind it.

The short version

Buying an agent platform puts your workflows inside someone else's product.

Read these first

What you get

Your brand
Set VITE_BRAND_PRODUCT_NAME and the other VITE_BRAND_* strings at build time, and override the colour tokens in src/brand/brand.css. The header, logo, document title and footer all carry your name. No core-logic fork.
Your cloud
One portable container with a ready-made deploy pack for each host: Docker Compose (laptop, VPS or on-prem, no cloud account), Fly.io, Render, Railway, AWS, Azure and Google Cloud.
Your model keys
Bring your own provider keys. The demo also seeds deterministic mock-AI workflows that run with no key at all, so you can evaluate before you spend anything.
Work you can stop and check
A step that would move money waits for a person to approve it. Every step is recorded as it happens, and a finished run can be rewound and played again without repeating its side effects, so a replay never pays twice.
A team to start from
Ten example agents (Ava, Felix, Cleo, Idris, Mira, Ren, Pax, Quill, Iris, Ezra), each with a board, cards and schedules. Rename them for your company, or ship a clean tenant with none.
A way out
The app speaks the open OpenWOP protocol, so your agents and workflows can move to any other host that implements it. Secrets never travel with them: you reconnect your keys where you arrive.

Get it running

  1. 1Download openwop-demo-app.zip from /install/ and check it against its published checksum.
  2. 2Start the whole stack on one machine with the Docker Compose pack: docker compose up --build in deploy/compose/. No cloud account needed.
  3. 3Put your brand on it with the VITE_BRAND_* settings and src/brand/brand.css, then rebuild the front-end.
  4. 4Rename the example agents for your teams, or turn demo seeding off for a clean tenant.
  5. 5When you are ready, deploy with the pack for your host under deploy/.

What you take on

  1. You operate itThere is no service contract or SLA behind the app, and no paid, managed offering from the OpenWOP project. Uptime, backups and incident response are yours, or your host vendor's.
  2. Upgrades are yours to scheduleThe bundle is rebuilt from the reference app as it changes. Taking a new version, and testing it against your branding and data, is a step you plan.
  3. Model costs are yoursWith your own provider keys, usage is billed to you by your provider. The mock-AI demo workflows cost nothing.
  4. The evidence is self-reported todayThe reference app passes the conformance suite, but every published result so far comes from the project's own steward or an affiliated organization. No independent organization has run its own host yet. For production evaluators explains how to read that.

The full guide

Everything on this page, in order, with every link.

The short version

Buying an agent platform puts your workflows inside someone else's product. Building your own means months spent on the parts every team needs: someone to approve a payment, a record of what happened, a way to undo a mistake. The third way is to start from an app that already works and make it yours. The OpenWOP reference app is open source (Apache 2.0), it is the same app that runs at app.openwop.dev, and it can carry your name, colours and logo without forking its core.

Read these first

What you get

  • Your brand. Set VITE_BRAND_PRODUCT_NAME and the other VITE_BRAND_* strings at build time, and override the colour tokens in src/brand/brand.css. The header, logo, document title and footer all carry your name. No core-logic fork.
  • Your cloud. One portable container with a ready-made deploy pack for each host: Docker Compose (laptop, VPS or on-prem, no cloud account), Fly.io, Render, Railway, AWS, Azure and Google Cloud.
  • Your model keys. Bring your own provider keys. The demo also seeds deterministic mock-AI workflows that run with no key at all, so you can evaluate before you spend anything.
  • Work you can stop and check. A step that would move money waits for a person to approve it. Every step is recorded as it happens, and a finished run can be rewound and played again without repeating its side effects, so a replay never pays twice.
  • A team to start from. Ten example agents (Ava, Felix, Cleo, Idris, Mira, Ren, Pax, Quill, Iris, Ezra), each with a board, cards and schedules. Rename them for your company, or ship a clean tenant with none.
  • A way out. The app speaks the open OpenWOP protocol, so your agents and workflows can move to any other host that implements it. Secrets never travel with them: you reconnect your keys where you arrive.

Get it running

  1. Download openwop-demo-app.zip from /install/ and check it against its published checksum.
  2. Start the whole stack on one machine with the Docker Compose pack: docker compose up --build in deploy/compose/. No cloud account needed.
  3. Put your brand on it with the VITE_BRAND_* settings and src/brand/brand.css, then rebuild the front-end.
  4. Rename the example agents for your teams, or turn demo seeding off for a clean tenant.
  5. When you are ready, deploy with the pack for your host under deploy/.

What you take on

  • You operate it. There is no service contract or SLA behind the app, and no paid, managed offering from the OpenWOP project. Uptime, backups and incident response are yours, or your host vendor's.
  • Upgrades are yours to schedule. The bundle is rebuilt from the reference app as it changes. Taking a new version, and testing it against your branding and data, is a step you plan.
  • Model costs are yours. With your own provider keys, usage is billed to you by your provider. The mock-AI demo workflows cost nothing.
  • The evidence is self-reported today. The reference app passes the conformance suite, but every published result so far comes from the project's own steward or an affiliated organization. No independent organization has run its own host yet. For production evaluators explains how to read that.

An app you own.