Understand the private style profile inferred from your own Flowgen projects, how it influences suggestions, and why an explicit request always wins.
Where: Flowgen Agent → a request where personal workflow or creative taste is relevant
Key ideas
- Private source: The preference document is derived only from the signed-in user’s own eligible flows and is marked private to that creator.
- Compact signals: It summarizes frequently used graph shapes, workflow vocabulary, and representative wording; generated media, binary files, and output URLs are not copied into it.
- Soft defaults: Preferences help when a brief leaves choices open. They must never override the current explicit instruction, required brand item, or newly selected reference.
Steps
- State the current outcome and any non-negotiable style, model, budget, or workflow requirements explicitly.
- Review the Agent’s proposed plan or inserted reference when personal preferences influenced an open-ended decision.
- Correct a preference directly in the current request when it no longer applies; the explicit instruction takes priority.
- Keep authoritative reusable facts in Brain Brand, Persona, or Business items rather than relying on inferred style.
- Inspect and adapt the graph before running, especially when the task differs from your historical work.
Preferences versus Brain
- Private preferences: inferred, compact, soft guidance about taste and workflow.
- Brain items: user-authored or intentionally written durable facts and structured context.
- Current prompt: highest-priority instruction for this task.
Privacy boundary
The preference record is scoped to its creator and injected only into that user’s retrieval context. Public Marketplace references are a different corpus source and do not make the private profile public.
Limitations and important notes
- A preference profile cannot guarantee a stylistic result; models and available tools still vary.
- There is no reason to place passwords, tokens, or private credentials in prompts or Brain items.