Usage scenarios

Valem is a runtime for state whose rules must always hold — expressed as data, not code. That shape shows up in more places than it first appears. These pages describe the four it fits best, with the concrete pieces each one uses.

Scenario The problem it solves Start here
Structured state for AI agents An agent that must keep a coherent, inspectable world model across a long session — and be able to prove why a value is what it is. MCP tools
Rules & calculations as data Pricing, quoting, eligibility, scoring: business arithmetic that changes far more often than your release cycle. Model guide
Effect-driven workflows State changes that must reach the outside world — call an API, ask an LLM, wait an hour — without losing determinism. Effects
Model-driven UIs A form or dashboard that should follow the model, not be rebuilt every time the model changes. View system

The common thread

In every one of them, the same three properties are doing the work:

  • Derived state can’t drift. You mutate base fields; the runtime recomputes everything that depends on them, in dependency order, on every change.
  • Invariants are enforced at the boundary. A rollback constraint refuses the mutation; a flag constraint records the violation. Neither is something a caller can forget to check.
  • Everything is explainable. Each derivation and constraint evaluation is traceable, and the optional durable audit trail is tamper-evident.

How they combine

The scenarios are not exclusive — a real deployment usually stacks them. A quoting service is rules-and-calculations at the core, with an http effect that pulls a live rate, a viewDefinition so the agent’s output is directly renderable, and an MCP surface so an agent can build and test new quote models against the same runtime.

Ready-to-run specs for all of this: the examples gallery.

Table of contents