AI Automation · Platform
Agentic AI Platform
Compose, govern and audit AI agents that read your documents, apply your rules and act inside the systems you already run.
What it is
Compose agents, then bound them
Building an agent is the easy half. Deciding what it may do without asking is the half that decides whether you can deploy it. The platform exists for the second half. The architecture underneath is on the Agentic AI Architecture page.
Compose
Named agents
Each agent owns one narrow job, so it can be tested, measured and replaced. An agent that does everything can be none of those things.
Compose
Workflow assembly
Agents are chained into a process you define, with the handovers explicit rather than implied by whoever built it.
Compose
Custom agents
Work specific to your business gets its own agent: your steps, your systems, your guardrails, defined in business terms.
Govern
Confidence thresholds
You set the score above which an agent may act alone. Below it, the work goes to a person with the specific field named.
Govern
Human gates
Value bands, document types or counterparties that must always have a person on them, enforced rather than encouraged.
Govern
Full decision record
Every action stored with the values seen, the rule applied and the outcome verified, queryable long after the fact.
Where it applies
What people run on it
A chain is a sequence of agents with a rule at each handover. These are the four shapes we see most. The first three come configured; the fourth is yours.
How PTAS AI does it
The record behind every action
A log that says the invoice posted is not much use in a review. What gets asked for is what the agent saw and which rule let it through, and the platform keeps both as a matter of course.
One decision, kept in full · Illustrative example
This is what separates an agent from an automation: it can tell you why, months later, without anyone reconstructing it.
Observed inputs
Agents watch their own queues rather than waiting to be called.
Reasoned decisions
Checked against your rules, not a general-purpose guess.
Verified outcomes
The agent confirms the downstream result, then closes.
The distinction
Agent, assistant, or a script
All three get called AI automation. Only one of them still owns the work when the format changes.
Three things often called the same thing
Repeats exactly what it was shown. Fast and reliable until the screen moves, the format changes or the path forks — then it stops, or worse, carries on wrongly.
Answers well when asked. It does not own the queue, does not act on your systems without a person driving, and carries no accountability for the outcome.
Watches its own inputs, reasons against your rules, acts inside your systems, records why, and escalates what it should not decide alone. The authority to act is the part that needs guardrails.
The useful question is not which is cleverest. It is which one still owns the work at 2am when the format changed.
Controls
Authority is configuration
The questions a CIO asks before switching this on are about limits, not models. These are the four we are asked most.
Authority boundaries
What an agent may decide alone is configuration, not a property of the model. Raise or lower it per workflow, per entity, per value band.
Audit and replay
Every decision carries its inputs and its reasoning, so a question six months later has an answer rather than a reconstruction.
Role-based access
Who can change an agent's rules, and who can only watch it run, are different permissions.
Data residency
Your region, your cloud, your retention period. On-premise and air-gapped deployment for regulated workloads, where supported.
The products
What runs on the platform
DocPro and FinSuite are both built on this runtime. You do not buy the platform on its own; you meet it through whichever product your work starts in.
DocPro
The document layer the agents read through.
Read more →FinSuite
The finance workflows those agents execute.
Read more →Custom Document Agents
A document type of yours, as a bounded agent on this runtime.
Read more →Enterprise Workflow Automation
Cross-system agent chains, described as a solution.
Read more →FAQ
Questions Teams Ask
What makes this an agent rather than automation with a model bolted on?
Three things together. It observes its own inputs rather than waiting to be invoked, it reasons against your rules rather than following a fixed path, and it has the authority to act inside your systems. The third one is why the guardrails matter. An assistant that is wrong wastes a minute; an agent that is wrong posts an entry.
How do we stop an agent doing something we did not intend?
By bounding what it may decide alone. Confidence thresholds, value bands, document types and counterparties can each require a person, and those boundaries are configuration you control rather than behaviour you hope for. Anything outside them escalates instead of proceeding.
Can we build our own agents on this?
Yes. Custom agents are part of standard onboarding. You describe the steps, the systems involved and the guardrails in business terms, and the agent is defined against those. Most customers end up with a mix of the standard finance and document agents plus a few of their own.
What happens when an agent gets something wrong?
Low confidence sends the item to review before anything is acted on, which catches most of it. Where something wrong gets through at high confidence, the decision record shows exactly which rule let it past, and that rule is where it gets tightened. The point of keeping the reasoning is that the fix is specific rather than a shrug at the model.
Is this a platform we licence, or something you run for us?
Either, and the choice is made with deployment. Some teams run it inside their own cloud or on their own hardware with their IT team holding the keys; others have us operate it. The agents, the controls and the decision record are the same in both cases.
Keep reading
Related Capabilities
AI Automation
The section this page belongs to, and the other five capabilities.
Read more →Autonomous AI Agents
The named agents that run on this platform, and what each owns.
Read more →Agentic AI Architecture
The technology page: how the runtime is built, for a CIO.
Read more →AI Orchestration
How several agents are sequenced, retried and supervised.
Read more →Agentic AI vs RPA
Where scripted bots break, where agents hold, and how to choose for a finance process.
Read more →Next step
Put one agent to work first
Pick the workflow that wastes the most time, bound what the agent may decide, and widen from there.