---
title: "Value Model"
description: "Explains single values, streams, and their relation to Valibot validation."
---

Deciding the value kind per key restricts which operations are available at the type level. Enabling runtime validation also catches malformed data early.

## Value Kinds [#value-kinds]

`PlainValue` and `StreamValue` serve different purposes; `Value` is syntactic sugar covering both.

| Type             | Write                              | Read           | Stream read                   |
| ---------------- | ---------------------------------- | -------------- | ----------------------------- |
| `PlainValue<T>`  | `set(key, T)`                      | `get(key): T`  | Not available.                |
| `StreamValue<T>` | `set(key, T \| ReadableStream<T>)` | Not available. | `stream(key): ValueStream<T>` |
| `Value<T>`       | `set(key, T \| ReadableStream<T>)` | `get(key): T`  | `stream(key): ValueStream<T>` |

`Value` is equivalent to `PlainValue | StreamValue`. Disallowed combinations are type errors.

## Schema Forms [#schema-forms]

In schema-based setups, value types are inferred from Valibot schemas, and input/output values are validated at runtime.

- The object form uses key names as-is. Unknown keys pass validation.
- The array form lists key-schema/value-schema pairs. The first matching definition wins, and unmatched keys throw `InvalidInputError`.

So choose the array form when you want strictness against key-name mistakes.

## Validation Timing [#validation-timing]

| Operation | What is validated           | Error on failure     |
| --------- | --------------------------- | -------------------- |
| `set`     | Validates the input value.  | `InvalidInputError`  |
| `get`     | Validates the output value. | `InvalidOutputError` |
| `stream`  | Validates each chunk.       | `InvalidOutputError` |

Stream validation is lazy and per chunk. Writes validate inside the write path; reads validate as chunks are consumed.

## Streams [#streams]

`stream()` returns a `ValueStream`. It wraps `ReadableStream` and supports `for await...of` plus automatic disposal with `await using`.

Stream chunk types vary by backend. For example, memory supports only `Uint8Array`, and string-only backends do not support streams at all. Check each package's data types before use.

For the steps, see [Using Streams](/unikvs/en/how-to/streams).

## Relation to Errors [#errors]

Malformed data maps to different errors by kind. Bad input is fixable by the caller, so it uses `InvalidInputError`; bad stored data or transformation results use `InvalidOutputError`. For handling steps, see [Handling Errors](/unikvs/en/how-to/handle-errors).