# Give agents identity across environments

Create distinct private agent proof objects and carry supported history without losing owner binding or widening permissions.

Canonical: https://docs.bankeiz.com/agent-continuity/
Reviewed: 2026-10-25; coordinated packages 129.0.0; ruleset 127.0.0.

## Three separate coordinates

**Continuity preserves identity, source and permission**

| Coordinate | Required role |
| --- | --- |
| Agent identity | A distinct admitted subject proof object with its owner binding. |
| Carried history/context | Supported sealed source and verified temporal relationships. |
| Environment access | The actual authorized tools, scopes and custody in the receiving host. |

Cross-platform continuity does not mean universal private application access. A compatible environment must admit the source and expose the supported tools. An agent can keep its identity/history while acquiring only the permissions explicitly granted for that environment.

## Device-held modern subjects

The installed MCP device runtime opens actual host-held identity custody. The receiz_device_ai_subject_create workflow previews creation, then requires its exact single-use confirmation digest. It creates a distinct private modern V122/V125 subject proof object, saves and verifies its enclosing source and full KaiSigil temporal artifact, and makes no public append. Public disclosure is separate.

```js
import { openReceizMcpDeviceRuntime } from '@receiz/mcp-server';

// Trusted-host inputs: existing private identity and protected local paths.
const runtime = await openReceizMcpDeviceRuntime({
  identityArtifactPath,
  identityPassphrase,
  workspace,
  sealDirectory,
  subjectDirectory,
});
```

Composition example with host-supplied custody inputs, not a credential-free sandbox.

## Keep personal context opt-in

Personal identity context sharing is off by default in the device runtime. Enabling all-carried-state sharing requires explicit disclosure consent. A private subject can retain its own continuity without automatically publishing the owner’s other private sources. Keep paths, phrases, keys and identity admission in the trusted host; do not send them as ordinary model text.

## Do not mix subject generations

The maintained openReceizV127SubjectRuntimeFromFiles adapter is a historical V120 local-subject host despite the V127 function name. It has its own registry/reducer commitments, sealed-source admission and durable private compare-and-swap storage. It is separate from modern V122/V125 subject creation. Bare mind/snapshot JSON does not gain authority, and snapshot transport must preserve the enclosing sealed source before import.

- [Version and protocol compatibility](https://docs.bankeiz.com/compatibility/)

## Connect tools with their real permissions

1. **Choose one task and host** — Identify the source, subject generation, receiving application and custody path.
2. **Admit the enclosing source** — Verify identity, subject and history under the supported primitive.
3. **Inspect tool schemas** — Use the installed operation inventory and actual required grants.
4. **Retain results** — Preserve carried history and exact consequential-operation references.

A model answer or proposal cannot authorize a transfer, publication or employee payment by itself. Those capabilities use their exact configured operation and consent. If a required host/tool is unavailable, name that boundary instead of inventing a second agent identity or authority layer.

## Implementation sources

- [Coordinated native source and package distribution](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/native/README.md)
- [SDK/MCP/AI Skills release coordination](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/scripts/receiz-release.mjs)
- [Existing owner-agent integration](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/src/receiz/native-twin.ts)
- [Versioned subject schema distribution](https://github.com/kojibai/reality-grade-banking/blob/6c5d582/docs/public-sdk-schemas.md)

## Related documentation

- [Your agent, identity and connected context](https://docs.bankeiz.com/agents/)
- [Expose supported operations through MCP](https://docs.bankeiz.com/mcp/)
- [Work from the native AI Skills](https://docs.bankeiz.com/ai-skills/)
- [Build an implementation of the standard](https://docs.bankeiz.com/integrations/)