Obsvra

Back to trackLesson 1 of 3

Agents 1: Free First, Pro When It Matters

The problem

An AI agent building a SpiceGrinder model has the whole component library to choose from — Free and Pro. Left unguided, it might reach for a Pro-only component out of convenience even when a Free-tier option is available. For a Free-tier user, how can you be sure the agent is recommending what you actually need, instead of nudging you toward an upgrade you don’t?

The rule isn’t a suggestion, it’s a documented instruction

The agent-facing skill provided by Obsvra that builds SpiceGrinder models from a description doesn’t leave this to judgment call — it’s written as an explicit instruction:

“…unless the user has confirmed they have Pro-tier access, prefer a Free-tier match and only recommend a Pro-only component when no Free-tier equivalent exists, explaining why the Pro-only one is necessary.”

That’s a testable claim, not a promise. You don’t have to trust that the agent is following it — you can check.

Case 1: Free actually covers it

Ask for claim severities, response times, order sizes — anything numeric with a shape (Normal, Lognormal, Poisson, Gamma, and 25+ others) — and every one of those distributions is Free tier. If an agent suggests Pro here, that’s a red flag you can catch yourself: open Component-Library-Reference.md, find the component it used, check the tag next to its name. ### Normal (Free). No Pro component was needed, so none should have been recommended.

Case 2: Pro is the only real answer

Ask for a synthetic SSN or EIN, though, and the honest answer changes: ### SSN (Pro), ### EIN (Pro) — and there’s no Free-tier equivalent to fall back to, because Free tier is capped to numeric-valued components only. A formatted, structured string like 123-45-6789 isn’t a number with different parameters — it’s a different category of output the Free tier doesn’t have a component for at all. Recommending Pro here isn’t an upsell, it’s the only component that produces that shape of data.

Similarly, refer to Modeling 101 Vital Signs lessons (1-3). These are demonstrate how a particular problem might appear on the surface to have a Free-tier solution, but refining the specification leads to needing Pro-tier features to meet the required fidelity.

What makes this verifiable instead of “trust me”

Three checkable facts, all inspectable independent of whatever the agent tells you:

  • Tier: every component’s Free/Pro status is public in the reference doc (or via the list_components MCP tool for a live, edition-accurate answer) — not something the agent asserts.
  • Numeric-only: Free tier can only produce numeric output. If your data needs a string, date, or formatted identifier, Pro isn’t optional — check whether the task itself requires non-numeric output, independent of which component gets suggested.
  • Node cap: Free tier stops at 20 nodes per model. If a model is small and numeric, there’s no legitimate reason for a Pro recommendation to show up at all. Even for a fully-numeric model, though, once it gets past 20 nodes, it will require upgrading to Pro.

None of these require taking the agent’s word for it. If a Pro recommendation doesn’t trace back to one of these three, that’s worth asking about.

Note that agents can and do make mistakes. If your agent using our skills suggests using Pro-tier features when Free-tier should be able to do the job, please let us know about it so we can try to correct and improve our skills package.