Knowledge and tests
Knowledge and Tests are top-level Build surfaces. Use them together: knowledge gives agents grounding material, and tests help catch regressions before Users do.
Knowledge
Section titled “Knowledge”The Knowledge page lists documents your agents can look up. The current dashboard describes supported source types as PDFs, DOCX, Markdown, and plain text.
Documents added from agent Knowledge tabs appear in the workspace Knowledge page when they are ready to browse. If the page is empty, add or link sources from an agent’s Knowledge tab.
Wallet-era workspace entitlements can affect storage and upload size. If the dashboard shows a storage limit message, review Billing before adding more content.
Agent knowledge attachments
Section titled “Agent knowledge attachments”Workspace knowledge is shared infrastructure, but each agent should have a curated set of attachments. Attach only sources that the agent should cite or reason from.
The Tests page has four tabs:
- Tests for reusable test definitions.
- Example sets for the examples a test can run against.
- Judges for scoring behavior.
- Runs for execution history and outcomes.
Use Tests to make changes safer. Build a test once, run it on meaningful agent changes, and inspect failed runs before distributing the agent.
Common recovery states
Section titled “Common recovery states”- Empty Knowledge means there are no browsable sources yet.
- Test run usage can be limited by wallet-era workspace entitlements.
- Failing tests should be treated as launch blockers until the prompt, knowledge, action, or judge expectation is corrected.