# Work that runs on its own, and stops for a person · Dscvry > Agents that plan their work, check every step and ask a person before they act, with a full trail of every run. Source: https://dscvryai.com/technology/agents/ # Work that runs on its own, and stops for a person. Multi-step work, within limits. It plans. It checks every step. It asks before it acts. The agents we build · How they work · Wherever the work is. Hosted Private cloud On-premise Edge Air-gapped ## The agents we build. Agents that do the work and stop for a person, inside limits you set and can check. - ### Checking Invoices against purchase orders, claims against contracts, reports against the rules. - ### Investigation Traces a break, a complaint or an incident across your systems, and writes up the evidence. - ### Watching Monitors feeds, regulations, contracts and deadlines, and raises only what matters. - ### Follow-through Chases missing documents, approvals and replies until the job is done. - ### Groundwork Assembles packs, drafts reports and fills in forms from your own records. - ### Routing Routes cases, updates records and hands each one to the right team. - ### Custom agents We design, build and run agents for the work your team does today, inside your boundary. ## Autonomy is easy. Accountability isn’t. An agent that can take actions across finance, operations and production systems is only useful if every action is bounded, checkable and stoppable. Most aren’t. ### Unbounded tools. Agents given broad access can call things nobody intended, with arguments nobody checked. ### No brake. Consequential steps like payments and postings happen without a person seeing them first. ### No trail. When something goes wrong, nobody can say exactly what the agent did, or why. ## How our agents work. Every agent works within the same controls, whatever the job. What will it do before it does it? ### A plan before any action. The agent writes down its steps first, so a person can see what it intends to do, and which step will need them. See it on purchase orders → https://dscvryai.com/use-cases/purchase-orders/ Could it touch something it shouldn’t? ### Only the tools it’s allowed. Every tool is typed and whitelisted per agent. Reading is not writing, and some actions are never available at all. See it on contract compliance → https://dscvryai.com/use-cases/contract-compliance/ Who approves the payment? ### A person signs the consequential step. Postings, payments and closures wait for a named person, who sees the evidence the agent used and can approve, edit or reject. See it on invoice errors → https://dscvryai.com/use-cases/invoice-errors/ Can we see exactly what it did? ### Every step, kept. Each plan, call, input, result, retry, refusal and approval is logged, so any run can be replayed and explained. See it on regulatory changes → https://dscvryai.com/use-cases/regulatory-changes/ ## Most of an agent is guardrail. Reasoning plans the work. Everything around it, from type checks and permissions to approvals and logs, is plain, testable code. Goal Understand A task with a clear outcome. Plan Reasoning Steps written before anything runs. Checks Rules & code Types, permissions and limits on every call. Approvals Human Consequential steps wait for a person. Trail Rules & code Every step recorded. ## The agent did the legwork. A person signed. Every run leaves a record of what it found, what it checked, what it stopped and who approved the rest. Here, a £4,500 reconciliation break: explained, drafted, approved by a person. **Tools used.** 4 read, 1 draft, 1 blocked write. **Retries.** 1, logged. **Could it have paid anyone?** No. send_payment is never available to this agent. Illustrative example. Accounts, payees and people are fictional. ## Built for regulated work. An agent should be easier to audit than the manual process it replaces. - ### Permissions per agent. Each agent gets exactly the tools its job needs, and nothing else. [How Operate works](https://dscvryai.com/technology/operate/) - ### Replayable runs. Any run can be replayed step by step from its trail. [Security & trust](https://dscvryai.com/security/) - ### Works in your systems. Agents act through the connectors you approve. [APIs](https://dscvryai.com/technology/apis/) ## Where agents help. Problems that take several steps across several systems. Does this PO match what we’re paying? ### Purchase orders Find mismatches before payment. See how → https://dscvryai.com/use-cases/purchase-orders/ What should we have charged? ### Revenue leakage Find money that should have been charged. See how → https://dscvryai.com/use-cases/revenue-leakage/ What actually happened with this complaint? ### Customer complaints Reconstruct what happened, and how it was handled. See how → https://dscvryai.com/use-cases/customer-complaints/ Which invoices are we overpaying? ### Invoice errors Catch overbilling, duplicates and mistakes before payment. See how → https://dscvryai.com/use-cases/invoice-errors/ ## Where it runs. Agents run where the work and its systems are, from hosted to air-gapped, acting only through the connectors you approve. ### A managed environment. Start with a hosted service when speed of adoption and managed operations are priorities. Agree data handling, access and service requirements as part of the engagement. Discuss hosted · ## Put them to work. Agents run on the rest of what we build. Each of these is built on the same tested foundations. ### Models Specialist models, evaluated on your data and run inside your boundary. Explore Models → https://dscvryai.com/technology/models/ ### Products Ready-made products for recurring jobs, from documents and contracts to sustainability data. Explore Products → https://dscvryai.com/technology/products/ ### APIs Model, agent and data APIs that plug into the systems you already run. Explore APIs → https://dscvryai.com/technology/apis/ ### Data Data collected, extracted and governed, with every value traced to its source. Explore Data → https://dscvryai.com/technology/data/ ### Evaluations Tests built from your own cases, rerun every time a model changes, so you always know how well it does your work. Explore Evaluations → https://dscvryai.com/technology/evaluations/ ## Bring us the work where wrong is expensive. Start with the operational problem. We’ll work out the data, models and agents it needs, and where it should run. [Talk to us →](https://dscvryai.com/contact/) **London**Discovery AI Limited · United Kingdom **Hyderabad**ReDscvry Technology Private Limited · India **Chennai**ReDscvry Technology Private Limited · India