Try it
Get started
From a checkout of the repository to your first initiative, on your own computer. You need Docker with Compose, Node.js 24.7 or later, and pnpm 12.
Ten steps
-
Install and start the database
In the repository's folder, install the packages and start PostgreSQL.
pnpm installgives this checkout its own database and its own address, and prints both.pnpm install docker compose up --waitProduces: A running database for this checkout.
-
Start the application
This builds what the server needs, applies the database migrations, and starts the application. The address it serves is in its log, on the line starting
serving.pnpm devProduces: The application at `http://pkw-<hash>.localhost:4000`.
-
Create your organization and admin
While no user exists, the log prints a one-time setup link (
no users exist yet: open …/setup#…). Open it, name your organization, and create the first admin with an email and a password of at least 15 characters. The page closes itself afterward.Produces: An organization and its admin, signed in.
-
Give yourself the product and reviewer roles
The first admin holds the admin role only, and creating an initiative needs the product role. Go to Administration, then Members, choose Change roles on your own row, tick Product and Reviewer, and choose Update roles. In a team, you would invite the product people and reviewers instead.
Produces: An account that can create initiatives and review knowledge.
-
Turn on the plugins
Go to Administration, then Plugins. Choose Enable… on Document upload, Manual entry, and Core workflows; each shows its settings, then the permissions it asks for under Review permissions, and is turned on with Enable. Core workflows provides the eight stage workflows.
Produces: Uploads, manual entries, and stages with workflows you can run.
-
Add a model key
Go to Administration, then Models and keys, add an organization key for OpenAI, Anthropic, or an OpenAI-compatible endpoint, and choose Probe to test it. Then choose a model for each task, or keep the defaults for your provider. Without a key you can still load and approve knowledge and create initiatives.
Produces: A provider the stage workflows can call.
-
Search by meaning on your own machine
Optional. Start the local embedding service beside the database; its first start downloads a small open embedding model. Then, under Administration, Models and keys, Embeddings and search, choose Use the local service. Search then matches meaning as well as words, and no passage is sent anywhere to be embedded.
docker compose --profile embeddings up --waitProduces: Search by meaning, served from your machine.
-
Bring in some knowledge
Under Knowledge, upload a few documents (Word, PDF, Markdown, text, or spreadsheets), or write a manual entry. Each one is processed and then waits for review.
Produces: Documents in the review queue.
-
Approve it
Open Knowledge, then Review queue, open a document, read what was extracted, and choose Approve under the decision on its version. For a source's first load, Approve all approves every document of that source, each with its own recorded decision. Approved documents appear in search.
Produces: Approved knowledge the workbench can cite.
-
Create your first initiative
Go to Workbench and choose Create initiative. Give it a title and describe the idea: who has the problem, what you want, and what is out of scope. Create it, and you land on its overview with eight stages. Start with problem framing.
Produces: Your first initiative.
Without Node.js
One command builds the application's container image, applies the migrations, and serves the application at http://localhost:4000. The setup link is in the container's log.
docker compose --profile full up --build --wait
What to read next
The guides live in the repository under docs/guides/: a user guide for reviewers and product people, an administrator guide, a reader's guide for engineering, technical guides for developers, and a presenter's script for each of the two fictional clients' demos. The repository's README covers configuration, keys and encryption, the local embedding service, and the tests. These steps follow the product as it is built today; where it stands says what is built and what is still coming.