How do I share an AI-built prototype with a customer?
Initialize the project, let the agent build the scoped workflow, then push, checkpoint and create a password-protected or invite-only share. Send the returned URL with specific review tasks.
Grain / Use cases / Send customers a working prototype, not a deck
Use case
A customer needs to react to a workflow, but a deck forces them to imagine the interaction and a localhost demo depends on your laptop. Grain gives an agent-built prototype a working URL, controlled sharing and comments that can drive the next revision.
Build a working onboarding prototype for Acme's operations lead. Create a three-step intake flow, a review screen and a confirmation state. Use clearly labeled sample data, add an assumptions panel, require no login, and collect no real customer information. Prepare a short task list for the reviewer.grain login
mkdir customer-onboarding-prototype
cd customer-onboarding-prototype
grain init
# Give the agent the scoped prototype prompt.
grain push
grain checkpoint -m "customer review one"
grain share --password <pw>Download the Mac app before running the sequence. For the sample onboarding flow, the pieces are a page with interactive states and a password-protected link. It uses no Postgres, login, scheduled job or custom domain because the prompt forbids real records and identity-specific behavior. If the customer must return to saved entries, promote the prototype to a full app and use Postgres. If each reviewer needs an identity, add login and invite them. If branding matters enough to pay for it, Pro supports a CNAME custom domain. Make each addition answer a testable need instead of treating infrastructure as polish.
Prove the prototype by testing the customer's task, not by counting generated screens. Open the link in a signed-out browser, enter the password, follow all three steps and verify that no real data is stored. Leave a comment on a confusing control and confirm that the agent can read it. Close the authoring session and reopen the same URL later. Grain's allowed product claims are that published artifacts stay published, readers can comment, and full apps may use Postgres, logins and scheduled jobs. Whether the concept is valuable remains a customer finding. Record what the customer could complete and where they stopped rather than inventing a conversion claim.
Initialize the project, let the agent build the scoped workflow, then push, checkpoint and create a password-protected or invite-only share. Send the returned URL with specific review tasks.
Usually no. Use labeled sample data. Add Postgres only when persistent records are necessary to test the workflow and the data is appropriate for the artifact.
Yes. Readers can leave comments on an artifact, and the agent can read and act on them in the next revision.
Yes. A full app artifact can have login. Sharing can also be restricted by password or invited people.
Yes. Pro supports a custom domain through CNAME. The default share URL works for the first review.
Download the free Mac app, scope one customer decision and publish the smallest honest interaction.
Free forever. No card. Bring your own agent.