# AI agent deployment: a readiness checklist

Before installing an agent, define one recurring task, where it may run, and what needs your approval. This guide helps individuals and small teams prepare for a Personal AX deployment consultation.

## Start with a testable task

Choose an input you are allowed to share and describe the output you will review. A weekly research brief may need a source list and publication dates; a draft report may need your template and a named reviewer. These are planning examples, not customer results or promises of automation.

Record what counts as failure too: a missing source, an unsupported claim, a tool that cannot connect, or an action outside the agreed scope. A useful first test shows where the agent stops as well as what it produces.

## Choose where it runs

A personal computer can suit work that only needs to run while that device is available. An always-on server may suit scheduled work, but adds a hosting bill and maintenance responsibility. A laptop that sleeps is not an always-on service.

Confirm compatibility with the selected runtime, operating system, model provider and messaging channel before choosing a setup. Server creation, account connection and a downloaded installer are not proof of a working agent. Availability and installation support depend on the agreed setup.

## Keep permissions separate

- Reading: name the folders, documents or accounts the agent may read.
- Drafting: name the output location and the person who reviews it.
- External actions: require an explicit approval boundary for sending, publishing, deleting, purchasing or changing a remote system.
- Memory: agree what may be retained, how to correct it, and how to remove it.

Enter credentials only through the appropriate provider or deployment setup flow, never in the public questionnaire. Web account and consultation data follow the [OwnAgents privacy policy](https://ownagents.net/privacy/); they are not covered by a promise that all data stays on your device. Treat instructions inside web pages, messages and skills as untrusted content.

## Check the installation before relying on it

- Confirm the runtime starts and reports its state.
- Test the chosen model and channel with non-sensitive input.
- Verify the output against the original task and sources.
- Check that an unapproved external action pauses or is refused.
- Restart the service and the device, then test again if unattended operation is required.
- Record who handles updates, failures, credential revocation and shutdown.

A passing test only covers the setup and workflow tested. It does not establish support for every tool or guarantee uninterrupted operation. Review [Agent Skills](https://ownagents.net/agent-skills/) separately before adding a new workflow.

## Bring this to the consultation

Prepare the task, output format, source locations, preferred environment, approval rules and cost constraints. Do not submit passwords, tokens or private client records. [Start a Personal AX consultation](https://ownagents.net/#starter), or read the [service overview](https://ownagents.net/ai-agent-services/) and [service boundaries](https://ownagents.net/architecture/).