Your SOPs, agreements and policies already say what the agent may do. Now it's enforced.
Proviguard reads your procedures, the policies behind each use case, your sponsor-bank program agreement and your data-processing terms clause by clause. Approval tables become the agent's limits, prohibitions become blocks, and deadlines become tracked obligations. Every rule stays tied to its clause and to the person who approved it.
Nothing is quietly skipped, and nothing pretends to be enforced.
Each clause is placed in one of five outcomes. Your compliance team sees the full list and decides what goes live.
Limits and approvers
An approval-authority table becomes the agent's limits: what it does alone, and who approves above each band.
Prohibitions and required steps
"Never…", "only after…", "only to…" become checks that hold or block the action before it runs.
Already covered
A clause your current rules already enforce is linked to them, not duplicated.
Deadlines and duties
Recorded against the document they came from, so your team can see and act on them.
Listed, not faked
Monthly reviews, retention and training don't happen when an agent acts. They're listed as a position you're taking, not a gap you didn't know about.
Hold your agents to what you signed.
Your sponsor bank's program agreement and your vendors' data-processing terms already decide what your agents may do. When an agent breaks them, it's your breach. Proviguard reads the clauses that apply at the moment an agent acts, and enforces them like any other rule.
- The bank sees its own clause behind each hold
- Fintechs stop learning about breaches in the next audit
- Customer data reaches only the vendors you contracted
- Your privacy team's answer, enforced on every call
Read. Test. Approve. Enforce. Trace.
Every clause, its outcome
From your document as written, with the exact sentence each rule came from.
On your real traffic
See what the rule would have done to last month's actions before it can stop anything.
Never on one person's word
Someone other than the author approves each rule. Every change is a new version.
Before the action runs
The agent's call runs, waits for the named approver, or is declined.
Every decision cites its clause
Asked why an agent was stopped, you show the sentence, the version and the approver.
A procedure only matters if the agent can't talk its way past it.
Giving the model your SOPs helps it do the work. It isn't a control you can rely on.
| SOP in a document | SOP in the prompt | SOP through Proviguard | |
|---|---|---|---|
| Does the agent follow it? | It can't read your wiki when it acts | Usually, until it reasons around it or is talked out of it | Checked before every action, outside the model |
| Who approved it? | Whoever wrote it | Whoever edited the prompt | Tested, then approved by someone other than the author |
| When it changes | Agents don't know | Someone edits every prompt | Changed clauses re-read and re-approved |
| When someone asks why | No link to what happened | "The model decided" | The clause, version and approver behind the decision |
In our testing, a rule written into the prompt held until one note in the case record said "pre-approved". Then every model we tested credited the money itself, in up to 30 of 30 runs. Enforced outside the model, the rule held every time. Read the research
The rules follow.
New version, new review
Only the changed clauses are read again. The new version is approved before its rules apply, and the old rules keep running until it is.
Expired means expired
Owners are told before an agreement ends. Rules that cite a document no longer in force are flagged for the team to resolve.
Past decisions stay explainable
A clause a past decision relied on is kept, so every decision can still be traced to the words it ran under.
Bring your own procedures, or start from ours.
The SOPs & agreements you already follow
Upload them, or sync them from where they live, starting with Notion. A document whose words change gets a new version, never a silent edit.
- SOPs and procedures
- Policies for any workflow your agents run
- Approval-authority tables
- Sponsor-bank program agreements
- Data-processing and vendor terms
A sample procedure to edit
Disputes, payment operations and agent-initiated payments each come with a sample SOP and sensible limits. Edit it, and it becomes your procedure and your agent's authority.
- Samples, not legal advice
- Your values, your approvers
The people who answer for the rules
- Compliance and risk: procedures enforced as written, with a clear list of what can't be
- Operations: change the SOP, approve the new version, and every agent follows it
- Legal: agreements confirmed before they apply, expiry tracked
- Engineering: business rules out of the agent code and prompts
Bring one SOP or agreement. We'll turn a clause into your agent's authority on the call.
A procedure, your program agreement, or one of our starters. You leave with the clause outcomes and a rule you can see run.