Some work comes back every morning. Read the supplier mail and file what arrived. Check each invoice against the rules. Ask someone when a number doesn’t add up. None of it is hard, and none of it is ever finished.
Automating it usually means a project — a specification, a developer, a release date — and afterwards, rules that the people who do the work can no longer read. So the routine stays manual, or it disappears into a black box.
Our new video guide shows another way. It runs seven and a half minutes, you can watch it at the top of this page, and it follows the accounts payable team of Keldmoor, a fictional distributor, through one week in DocReveal’s Agent Manager. The team builds two agents — one files supplier mail, the other checks every invoice — tests them, publishes them and keeps every run under its control. The idea behind it is in Agent Pool; here is how it looks in practice.
A helper for routine work
In DocReveal, an agent is a helper for routine work: instructions in plain words, a set of tools and a budget.
It all happens in the Agent Manager, on five tabs: Overview, Agents, Studio, Orchestration and Activity. No code, no IT project: the people who know the work set it up, and every run is budgeted and audited.
Your first agent, guided
With no agents yet, the Agents tab offers a guided first agent, or a blueprint — a ready-made agent for a common job. Keldmoor’s choice: read mail and file documents. The guide asks for the mailbox the agent watches (the accounts payable mailbox, already set up in Connection Hub), the folder, the senders it may read — Keldmoor’s suppliers — and how often to check: every five minutes.
The test runs the agent on a sample message as a plan, so nothing is written. Once its test cases pass, the agent goes live and files supplier documents into DocReveal on its own.
Describe the job in plain words
For invoices, the team opens the agent wizard: six steps from a blueprint to a published agent. The Invoice triage blueprint reacts to every validated document — it approves clean invoices and escalates failed or suspicious ones.
The instructions are the agent’s job description in plain words: read the validation result, let a passed invoice through, hold a failed one with a label and ask a person what to do next. Every save is a new version, and the Diff view shows exactly what changed.
Decide what it may do and when it runs
The triage agent may use ten tools: read the document, its validation result and the mapped fields, ask a person, remember, and assign labels. More come from DocReveal, your connectors and the built-in set, and work can also be handed to other agents and flows.
Tools that only read always run. Tools that change data are marked sensitive, and their group’s policy decides: run automatically, ask for confirmation, or wait for a person’s approval.
Triggers start the agent on its own. This trigger reacts to every document that has finished validation; others run on a schedule, watch a mailbox or start the agent on demand. One switch lets external applications that hold your API key find and run the agent — left off, it stays internal. The summary checks the definition against your tariff, and publishing waits until the agent passes its eval set.
Try it before anyone relies on it
In the Studio, the team chats with the draft. The run input plays the event the agent will receive: here, a second copy of an invoice that is already in DocReveal. The agent reads the validation result, the mapped fields and the document profile, and every tool call is listed.
One rule failed — the invoice number must be unique per supplier. The agent won’t approve a duplicate; it asks instead. After the answer, it lays out its plan: hold the copy with a label and remember the case. Nothing is written, because this is a test. A test run does the same unattended, and the trace records every step, including the wait for your answer.
No GO, no publish
On Wednesday, the team sharpens the instructions. The draft has changed, so publishing is blocked until the eval set passes again.
The eval set runs the agent against its test cases — from the blueprint, plus your own if you add them. Hard checks are exact: the right tools, the right result. Soft checks are graded by a model against a rubric. GO: both cases pass, and publishing unlocks for exactly this draft. One more look at the summary, one click, and version three goes live.
It asks, you decide
On Thursday afternoon, a Stavlund invoice arrives. Its line items add up to $8,240; the printed total says $8,420. The agent holds it and asks, and the question waits in Activity, under Waiting for approval.
Open the run, and the question comes with ready answers. Pick one, and the agent continues where it stopped: the invoice is escalated and labeled “Do not pay – disputed”. Your decision is on record. Agent runs lists every run of every agent, and Recent activity is the audit of what they did.
The same inbox also collects approvals for an external AI assistant connected to DocReveal — the story of Agents under control.
It learns what you confirm
The same run proposed a note for next time: Stavlund invoices that don’t add up are held, and a corrected invoice is requested. Nothing enters the agent’s memory until a person confirms it; after that, the note is part of every future run. Run journals record what each run did: the tools, the decision, the question.
Agents that work together
Orchestration maps how agents and flows work together. At Keldmoor, the validated event starts the triage agent, and a flow sends approved invoices to the ERP. Click an agent to add a delegation — work handed to another agent. Chains stop after two hops.
Your rules, your limits
The Overview shows your tariff’s limits and what you have used: runs, tokens and the autonomy ceiling. The brain card is a short company profile that every agent receives, generated from your documents or written by you.
An agent can be paused with one click and resumed the same way. It can be exported and imported as a new draft; secrets never leave, and triggers arrive switched off. In Settings, the Agent Manager section holds the defaults for autonomy, budgets, memory and notifications.
One thing to know: building agents and answering their questions takes the administrator role. The rest of the team can follow every run.
What to take away
- The people who know the work build the agent. A blueprint to start from, instructions in plain words, no code.
- Nothing goes live untested. The Studio shows every step, and publishing waits for a GO from the eval set.
- The agent asks when it matters. Questions wait in one inbox, and every answer is on record.
- You stay in control. Tool policies, limits, pause and confirmed memory decide how far an agent goes.
Keldmoor and its suppliers are fictional demo data; the screens are the real product.
Hand it over. Keep the final say.