Tectonic Agents — Hermes-powered learning agents for Minecraft & Discord

Customer Docs

Models & Weights

Choose by task fit, not by provider economics. Light, Standard, Heavy, and Max are customer-safe capability labels for a platform-managed, versioned model catalog and routing policy.

A weight is a product label, not a model-quality guarantee, SLA, provider name, or published cost multiplier. The catalog and routing policy are managed and versioned by Tectonic.

The four task-fit labels

Light

Fast, focused work where latency and volume matter most.

Typical fit: Simple support answers, classification, routing, and short explanations.

Standard

The balanced default for everyday tenant work.

Typical fit: Normal orchestration, grounded support, and routine multi-step questions.

Heavy

Deeper reasoning for longer context or more interacting constraints.

Typical fit: Crash investigation, compatibility analysis, and careful code or configuration review.

Max

Highest-scrutiny routing for the hardest or highest-risk cases.

Typical fit: Rare escalation work, difficult debugging, and acceptance-sensitive review.

How the catalog and routing work

The platform publishes the managed catalog and assigns the task-fit label used in the customer UI. Routing can select a platform-managed default or a versioned alternative when the task, safety policy, availability, or account configuration calls for it. The exact underlying model identity is an implementation detail that may change through a catalog/configuration update; a label is not a permanent provider commitment.

Where the Agents surface allows a tenant to choose a model, the choice is made from the current catalog rather than from an arbitrary raw provider string. Custom-provider entries may appear without a platform weight because Tectonic cannot make the same managed-catalog comparison for an endpoint it does not operate.

How weights relate to credits

Heavier task-fit labels generally indicate more capable or more expensive-to-run work, so a long Heavy or Max task may use more credits than a short Light task. The actual debit is determined by the released metering and entitlement rules for the action; the label itself is not a public multiplier and does not replace the credit ledger.

Use Light or Standard for high-volume support and routing. Reserve Heavy or Max for diagnostics, code, and review work where additional reasoning is worth the extra bounded spend. This is guidance, not a promise that every role or route is enabled in every account.

What a weight does not mean

It is not a raw provider price, Routera multiplier, margin disclosure, or pass-through COGS statement.
It is not a guarantee that every task will use the same model, context window, latency, or answer quality.
It does not grant observe, propose, or act authority. Trust and effector gates remain separate from model selection.
It does not make a fixture, synthetic showcase, or local catalog listing proof of a production route.

Platform-managed and versioned

Catalog entries, default routes, task-fit labels, and their billing bindings should be treated as versioned platform configuration. When a route changes, the customer-facing label and applicable account semantics should change together. Until a catalog entry is present in the released account surface and its path is verified, it is not a promise of availability.

Credits & Billing