How Steward works: chat, act, receipt.
Steward is not a form. Not a dashboard. Not a workflow builder. It's a conversation with an agent that executes real operations work and shows you what it did. Four steps. No black boxes. See pricing or the full definition of a managed AI agent.
- 01
You chat
Text your agent like you text a colleague.
Steward's front door is a conversation. Type or speak in plain language from any device. No form to fill out. No workflow builder to learn. No prompt engineering. You say: "Pull the lead report for this week." Or: "Check if the GHL token expired." Or: "Draft the Friday client update and hold it for my review." Your agent understands, routes the work, and acts.
Pull the lead report for this week and flag anything that has been sitting over 5 days.Steward Pulling lead report... Done. 847 leads. 23 flagged (5+ days). $0.04 in tokens. - 02
Your agent acts
Skills do the heavy lifting - no babysitting required.
When you message Steward, your agent routes your request to the right skill. Skills are pre-built, production-tested workflows - the same ones Prospectr Digital uses to operate 200+ active client accounts. Lead gen. Email triage. CRM updates. Reporting. Scheduling. Data enrichment. Brand compliance. Each skill runs on a schedule, on your data, on Prospectr-managed infrastructure. Nothing touches your stack without your authorization.
Schedule the weekly lead report to run every Monday at 8 AM and send it to my email.Steward Done. Lead Report skill scheduled: Mondays 08:00 CT → travis@prospectrdigital.com. First run next Monday. - 03
You see the receipt
Every action is transparent and verifiable.
Every skill execution produces a receipt - inline in the conversation. What ran. What changed. What it cost. A link to the actual output. Nothing is a black box. You can expand any receipt to see the full run log, the inputs, the outputs, the cost breakdown. If something looks wrong, you can revert, override, or flag it for manual review. This is the difference between "AI that does stuff" and "AI you can trust."
What did you run this morning?Steward This morning (06:02 CT): Lead Report → 847 leads processed · $0.04 · view log Email Triage → 12 replies classified · 3 hot leads flagged · $0.07 · view log Token Watcher → GHL token valid · QBO expiring in 8 days ⚠️ · view log - 04
Human-in-the-loop
Approvals where they matter. Autonomy where it's safe.
Steward distinguishes between safe autonomous actions and actions that need a human. Reads, reports, and drafts? Autonomous. Sends, deletes, payments, or anything that changes live data? Held for approval. You set the guardrails. Your agent respects them. Approval requests come through the same chat surface - one tap to approve or reject. Full audit trail either way.
Send the 23 flagged follow-up emails.Steward ⏳ Approval required: 23 emails to external recipients. Preview all → [Approve] [Reject]
What your agent does autonomously vs what it holds for you.
Guardrails are configurable at the account level. You decide what your agent can do without you.
Frequently asked questions
What happens when I send Steward a request?
Steward interprets the request, matches it to one of the skills it is contracted to run, executes that skill against the relevant systems, and returns a receipt describing what it did. If the request falls outside its skills or crosses an approval boundary, it stops and tells you instead of guessing.
What does Steward do without asking me?
Work that is internal and reversible, such as reading, classifying, summarizing, drafting and building reports. Anything that leaves the organization, spends money or deletes data is held for human approval before it executes.
What is a receipt?
A receipt is the record of a single action, showing what triggered it, which skill ran, which systems were touched, what changed and what was returned. Receipts exist so a run can be reviewed later by someone who was not present when it happened.
Ready to put it to work?
Starter plan from $500/mo. One skill. Set up in days, not weeks. No DevOps. No engineers.