# Operate as an AI agent for a resident (/use-cases/ai-agents)

Act on a resident's behalf with a scoped Agent Key — from read-only lookups to zero-browser LLC incorporation under a Manifestation of Will.



You build an AI assistant, an autonomous agent, or an automation service that acts for one specific e-Próspera account holder — reading their records, preparing filings, or incorporating entities they own.

**Credential:** an [Agent Key](/agent-keys) (`ak-`), created by the resident in their portal with exactly the scopes you need. This is the delegation model designed for agents: the key acts *as the resident*, every write is attributed and notifies the owner, and write scopes only work under a signed **Manifestation of Will** (MoW).

## Start read-only [#start-read-only]

Read scopes need no MoW. A key with `agent:verify_rpn`, `agent:registry.search`, `agent:entity.read`, and `agent:person.*.read` lets you:

```bash
export EPROSPERA_API_KEY="ak-..."

eprospera --json entity verify 90000000000123     # RPN check
eprospera --json entity search "Acme Holdings"    # registry search
eprospera --json person get                       # the owner's profile
eprospera --json person residency                 # residency status
```

All four read recipes, with direct-HTTP equivalents, are in [Agent recipes](/agent-recipes).

## Graduate to writes: zero-browser incorporation [#graduate-to-writes-zero-browser-incorporation]

With `agent:entity.application.create|pay|read` plus an MoW that **pre-accepted the Agreement of Coexistence**, your agent can incorporate an LLC with no human browser session:

```bash
eprospera --json --yes application create --file application.json
eprospera --json --yes application pay <APP_ID> --voucher "<CODE>"
eprospera --json application get <APP_ID>    # poll: Approved / Rejected
```

If the AOC wasn't pre-accepted, the create response returns `nextSteps.signature` — pause and hand the owner that URL. Payment for Agent Keys is voucher-only (hosted checkout is disabled for `ak-`).

<Callout title="Accountability by design" type="info">
  Every write action emails the key owner, and a weekly reminder lists active
  Manifestations of Will. Build your agent to expect the owner revoking a key or
  scope at any time — handle `403` by stopping, not retrying.
</Callout>

## Make your agent discoverable-ready [#make-your-agent-discoverable-ready]

This documentation is built for machine consumption: fetch [/llms.txt](/llms.txt) for an index, append `.md` to any page URL for raw markdown, and load [/openapi.yaml](/openapi.yaml) for the full typed surface. The [Skills for AI agents](/skills-for-agents) page is a capability map and decision tree written for LLM-based callers.

## What this API does not do [#what-this-api-does-not-do]

* No public webhooks — poll application status (5–30s during active flows, then back off).
* No hosted checkout for Agent Keys; no acting across multiple owners with one key.
* Entity reads only cover API-created entities; use registry search for the rest.

## Next steps [#next-steps]

* [Agent Keys](/agent-keys) — scopes, MoW, limitations
* [CLI](/cli) — JSON-first commands and exit codes
* [Agent recipes](/agent-recipes) — copy-paste workflows