Action control for AI agents
Your agents can reason. Gatewise decides if they can act.
Gatewise checks every agent action against deterministic business rules, real system data and approval policies before anything happens.
Works with your agents and existing business tools.
- ALLOW
- BLOCK
- REQUIRE APPROVAL
- NO ACTION
- INSUFFICIENT EVIDENCE
Action request
refund.create
Refund 349 EUR to Acme Corp
waiting for decision
Evidence and policy check
- Invoice statuspaid
- Customer verifiedyes
- Existing refundnone
- Automatic refund limit100 EUR
- Manager approval threshold100 - 500 EUR
Result
evaluating...
What Gatewise is
Gatewise is an action control layer for AI agents. It evaluates proposed business actions against deterministic policies and trusted evidence before execution.
The gap
Agents are becoming operators. Their controls haven't caught up.
An agent can have permission to call a tool and still lack the business context to know whether the action is appropriate. Tool access is static. Business rules depend on state, amounts, timing and who is affected.
- 2,000 EUR refundPayments
- 25 percent sales discountCRM
- Customer contract updateContracts
- Account deletionInternal tools
- Outbound email campaignSupport / marketing
- Renewal escalationCRM
Authorization
"Can this agent access this tool?"
Answered once, at setup. A token, a scope, a role. It cannot tell a 20 EUR refund from a 20,000 EUR refund.
Gatewise
"Should this specific action happen right now?"
Answered on every request, using current system data, thresholds, duplicates and the policy your business already follows.
Business rules should not depend on an LLM remembering them.
How it works
Three steps between intent and execution
- 01
Intercept the action
An agent sends an action request through Gatewise before executing it. Nothing reaches the business tool yet.
- 02
Check the evidence
Gatewise evaluates trusted business data, policy rules, thresholds, duplicates and approval requirements.
- 03
Return a decision
Gatewise returns one of five deterministic outcomes with a reason. The agent continues only when the policy permits it.
The five decisions
- ALLOW
Evidence and policy conditions are met. The agent may execute the action.
- BLOCK
A policy rule forbids this action. The agent cannot execute it.
- REQUIRE APPROVAL
The action is valid but above an autonomous threshold. A named human has to approve it.
- NO ACTION
The action is unnecessary. A duplicate or an existing task already covers it.
- INSUFFICIENT EVIDENCE
Required system data is missing or stale, so the decision cannot be made safely.
Control center
Every controlled action, with the reason attached
Operations teams see what agents proposed, what was allowed, what was stopped and what is waiting for a person. Blocked actions are the ones that tell you where your agents are drifting.
- 09:41Refund 79 EURsupport-agent-v3ALLOW
- 09:43Discount 25 percentReason: Exceeds 10 percent autonomous discount limitsales-copilotBLOCK
- 09:44Renewal follow-upReason: Existing task already assignedcs-agentNO ACTION
- 09:47Refund 640 EURReason: Finance approval requiredsupport-agent-v3REQUIRE APPROVAL
- 09:51Delete customer recordReason: Deletion restricted to humansops-agentBLOCK
Interface example. Not Gatewise customer data.
Policy
Policies in business language, enforced as code
A refund rule is a short list of statements the finance lead can read and approve. Gatewise turns business policy into deterministic execution controls, versioned and owned by your team.
- R01Allow refunds up to 100 EUR automatically.
- R02Require manager approval from 100 EUR to 500 EUR.
- R03Require Finance approval above 500 EUR.
- R04Never issue a refund when a chargeback or previous refund already exists.
- R05Require invoice status = paid.
Interface example of a configured policy.
Adoption
Start in shadow mode. Block nothing on day one.
Run Gatewise without enforcement. It observes agent actions, evaluates them against your policies and shows what would have been allowed, blocked or escalated. You turn enforcement on per action type, when the numbers make sense.
- No change to agent behaviour while you measure.
- Policies get tuned against real traffic, not assumptions.
- Enforcement is enabled action type by action type.
- actions evaluated
- 1,284
- would have been blocked
- 42
- required approval
- 11
- duplicate actions detected
- 7
Interface example. These numbers are not Gatewise customer results.
Use cases
The actions companies control first
Every use case starts from a business rule that already exists for human staff.
AI agent refund controls
Control autonomous refunds using transaction state, customer data and monetary thresholds. Gatewise decides whether a refund may be issued before the payment tool is called.
View controlsAI sales discount controls
Enforce discount limits and route exceptions for approval. Gatewise checks deal size, margin rules and prior concessions before a discount is applied.
View controlsAI customer communication controls
Prevent duplicate, inappropriate or unauthorized customer outreach. Gatewise checks contact state, consent and recent history before a message leaves your systems.
View controlsAI CRM action controls
Control record updates, ownership changes, lifecycle changes and destructive actions. Gatewise separates safe enrichment from changes that move revenue data.
View controlsAI contract and renewal controls
Verify timing, ownership and business rules before renewal related actions are executed. Contract terms are precise, so the controls around them should be too.
View controlsAI finance workflow controls
Apply deterministic evidence and approval rules before financially sensitive actions. Payouts, credits and write-offs get the same treatment as human initiated finance work.
View controls
Works with your stack
Works with the tools your agents already use.
Gatewise is designed to sit between agent runtimes and business systems, so actions can be checked before they execute.
Agents and AI
- OpenAI
- Anthropic
- Microsoft Copilot
- Google Gemini
- LangChain
- CrewAI
- Vercel AI SDK
Gatewise
Policy + evidence + approval check
- Evidence
- Policy
- Thresholds
- Approval
Business systems
- Salesforce
- HubSpot
- Stripe
- Zendesk
- Intercom
- Slack
- ServiceNow
- Shopify
- Microsoft Dynamics 365
- Jira
Designed to work across your agent and business stack.
Using APIs, tool calls and agent gateways. More integrations are being added.
All names and logos are trademarks of their respective owners, shown for identification only.
Architecture
A decision layer, not another agent framework
Gatewise is agent-framework agnostic. It is designed to sit between agent runtimes and the tools they use, so it applies to categories such as CRM, support, payments and internal tools rather than to one vendor.
- proposes
Agent / copilot
Any agent runtime or framework. Gatewise does not care which one.
- intercepts
Gatewise API / SDK / gateway
Designed to sit between agent runtimes and the tools they use.
- decides
Policy and evidence engine
Deterministic rules, thresholds, duplicate checks and required evidence.
- executes
Business systems
CRM, support, payments, billing and internal tools.
Auditability
Every decision comes with a reason
Gatewise records proposed actions, not only executed ones. That record is useful for operations, for compliance reviews and for debugging an agent that keeps proposing the wrong thing.
- Requested action
- Agent
- Policy evaluated
- Evidence used
- Decision
- Reason
- Approval
- Execution status
- Timestamp
Questions
Frequently asked questions
Short answers about action control, guardrails, approvals and audit.
Give your agents freedom without giving them a blank cheque.
Tell us what your agents are allowed to do today. We will review whether Gatewise fits your agent workflow.
Works with your agents and existing business tools