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

# Move to @agentium/harness

> Replace removed Agent-owned harness configuration with explicit package definitions, drivers, and runtime ownership.

The harness is a separate package. Import its runtime, drivers, definitions, abilities, and policies from `@agentium/harness`. Continue importing `Agent`, `Team`, `Workflow`, providers, and the neutral execution contracts from `@agentium/core`.

<Note>
  These removals apply in Agentium 4.0.0. Upgrade core and harness together and follow the [full v4 migration guide](/migration-v4) for other integrations.
</Note>

## Replace removed entrypoints

| Previous API or pattern | Current approach |
| - | - |
| `Agent.deep()` | Explicit `HarnessRuntime` with `agentDriver(config)` and the capabilities your application needs. |
| `new Agent({ harness, harnessOptions })` | Supply a definition and runtime configuration to `HarnessRuntime`; bridge the Agent through a driver. |
| `Agent.replaceTools()` | Compose runtime tools/abilities before a run and narrow active tools through a controller. |
| Agent lifecycle implicitly owning a harness | Host constructs the runtime, supplies identity/grants, and owns session-resource cleanup. |
| Assuming Agent storage is harness history | Use `HarnessSessionStore`; import legacy display transcripts explicitly when appropriate. |
| Implicit deep-agent filesystem/skills access | Configure defaults explicitly and grant the intended tool names. |

The package quickstart shows the [complete replacement](/harness/quickstart). A minimal migration keeps your Agent configuration, wraps it with `agentDriver(config)`, and starts it through `runtime.run(input, { identity, sessionId })`.

## Decide ownership during migration

Passing an existing Agent borrows it. Passing configuration creates a run-owned Agent and applies definition defaults/limits. If your old setup relied on implicit defaults, enumerate workspace access, context files, skills, filesystem, subagents, file memory, and session search before enabling them.

Move execution permissions to `grants`, run-wide limits to `budgets`, and step selection to a controller. Keep host authentication outside the harness. Do not assume an old transcript contains enough information to resume a partially executed run; legacy imports are marked non-replayable.

The old `/agents/harness` and `/api-reference/harness` URLs redirect to the package overview and complete reference. They no longer identify separate Agent features.

Next: [definitions](/harness/definitions), [runtime configuration](/harness/runtime), or [API reference](/harness/api/overview).


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