Distinguish reusable files from run inputs and request an output downstream modules can validate.
Workflow & automation · Beta / preview · Documentation checked 2026-10-04
Make scenarios and Make AI Agents (New); connected model providers have their own terms.
Scenario limits vary by plan. AI Agents (New) is in open beta; custom AI providers require a paid plan.
The worked example is illustrative and has not been tested in a signed-in product account.
What it does and how it connects
Distinguish reusable files from run inputs and request an output downstream modules can validate.
Knowledge files provide reusable agent reference material; Input files belong to a run. A conversation ID can reuse a thread, while a blank ID starts a new one. Structured output helps downstream mapping, but valid shape does not establish factual correctness.
Walkthrough
Use the product named in the access panel. The steps below combine documented controls with the editorial example.
- Add a small approved reference as Knowledge and keep run-specific documents in Input files.
- Choose a conversation-ID strategy scoped to the intended user or task.
- Define the supported response structure with fields needed downstream.
- Test known, unknown, and contradictory cases and validate the returned fields separately.
Worked example
Classify a fictional workshop request into topic, missing_details, and needs_review. Keep different requesters in different conversations. The original text remains the evidence against which the classification is checked.
Classify this fictional workshop request. Return topic, missing_details, and needs_review. Use only the approved topic guide. Keep missing details explicit; do not infer a person’s identity or consent. If the guide does not cover the request, mark needs_review. No external actions.
Inspect and verify the result
Check field shape and allowed values in a downstream validation step. Compare content with the original request and reference guide. Test that unrelated requesters do not share a conversation ID.
Limits and troubleshooting
Structured output can be syntactically correct but unsupported. Excessive history or stale Knowledge can mislead the agent. Update approved references deliberately and use a fresh conversation when contexts should remain separate.
Data, permissions, and actions
Knowledge and conversation history can persist across runs. Run files may contain personal data. Scope IDs and sources carefully and review provider and platform retention; never reuse a global conversation ID for unrelated users.
Related reading
Credits, provider billing, and run review
Track scenario usage and model-provider usage as different parts of operating cost.
Fixed AI triage with a review filter
Classify synthetic requests while keeping authoritative review separate from AI output.
A bounded beta knowledge agent
Use the beta agent to retrieve approved workshop evidence and produce a validated response draft.
Permissions and meaningful approvals
Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Make AI Agents (New) configuration · Make
AI Agent best practices · Make