> ## Documentation Index
> Fetch the complete documentation index at: https://vetta.sh/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Memory

> Mounted, versioned documents that persist across sessions.

> [!NOTE]
>
> **Phase 3.** Memory ships alongside [multiagent orchestration](/docs/team/overview.md).

Memory gives an agent knowledge that outlives a single [session](/docs/concepts/sessions.md). A memory store is a set of **versioned documents** you mount into an agent; the agent reads them for context and can propose updates, which you version like any other document.

## Memory stores

A memory store is an independent, organization-scoped resource — like [skills](/docs/capabilities/skills.md), it is unaffected by session deletion. The difference: skills are authored expertise you push; memory is knowledge that accrues from the agent's work.

```bash CLI theme={"system"}
vetta memory create --name customer-facts
vetta memory put customer-facts --key acme --file ./notes/acme.md
vetta memory list customer-facts
```

## Mounting into an agent

```typescript TypeScript theme={"system"}
const agent = await vetta.agents.create({
  name: "account-manager",
  model: "zai-org/GLM-5.2-FP8",
  harness: "pi",
  budget: { capUsd: 50, maxTaskUsd: 5, period: "month" },
  memory: ["customer-facts"],
});
```

Mounted documents load with the same progressive-disclosure discipline as skills, so a large memory store never floods the context window — the agent pulls only what a task needs.

## Versioning

Every write creates a new version. An agent's proposed update is a version you can review, so memory is auditable rather than a silent mutation.

> **[Next: outcomes & graders](/docs/capabilities/outcomes.md)**
>
> Score a session against a rubric.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.