Direct access creates blind spots.
- Permissions are scattered across tools
- Sensitive actions are difficult to review
- Decisions and outcomes are hard to trace
THE CONTROL PLANE FOR AGENTIC AI
Give AI
freedom.
Keep control.
Put a policy layer between AI agents and your systems.
Govern every tool call—with identity, permissions, human approvals, credentials, and audit built in.
Your agents can act. Your policies decide how.
+ INTELLIGENCE, WITH INTENTION. +
✳
Your AI agentReady to take action
••• Identity verified Policy checked
custos ONE GOVERNED GATEWAY
Read customerAllowed
Create refundNeeds approval
Delete recordsBlocked
Every decision. Accounted for.Recorded in your workspace audit trail
ILLUSTRATIVE WORKFLOW 01 — 03
Built around the tools
you already have.
MCP servers
OpenAPI
REST APIs
One consistent layerof control.
FROM DIRECT ACCESS TO GOVERNED ACTION
AI agents are moving
from answers to action.
As agents touch more systems, access control, approvals, and accountability become part of the product—not an afterthought.
Every tool call follows your rules.
- One gateway for agents and enterprise systems
- Policies and human approval before execution
- Identity, credentials, and audit in one path
THE CONTROL PLANE FOR AGENTIC AI
Big possibilities.
Clear boundaries.
Bring your tools together. Decide what agents can access and do.
Keep every action accountable.
01 / CONNECT
Every tool, in its place.
Connect services, discover their tools, and review new versions before publishing them to your teams.
Meet your gateway ↗02 / GOVERN
Access with intention.
Set permissions by role, user, tool, and arguments. Keep sensitive actions in the hands of a human reviewer.
Try a policy in action ↗03 / UNDERSTAND
Clarity at every step.
Follow requests, policy decisions, and outcomes in one audit trail. Understand usage and export the records you need.
Explore the controls ↗A SMALL LAYER. A BIG DIFFERENCE.
Between “can it?”
and “should it?”
AI agents can call tools. Custos decides whether they should. Every request follows the same governed path, so your rules follow every action.
Follow a request- 01
Connect & review
Bring in MCP servers, APIs, and internal services. Inspect discovered tools and publish the ones your team needs.
- 02
Set your boundaries
Define access policies, scope credentials, and choose which actions need approval.
- 03
Let good work happen
Agents call the gateway. Custos checks the request, governs execution, and records the outcome.
SEE CUSTOS IN ACTION
Governance you can
see in motion.
Watch how Custos puts a clear control layer between AI agents and the tools they use.
LESS GUESSWORK. MORE GOVERNANCE.
See Custos
make the decision.
Choose a real-world tool call and follow identity, policy, approval, execution, and audit in one path.
INTERACTIVE DEMO · SAMPLE DATA
01 — CHOOSE A REQUEST
PRINCIPAL support-agentROLE USER
02 — FOLLOW THE DECISION
tools/call
read_customer({ id: "customer_1042" })↗- Verify identityWorkspace membership & roleReady
- Evaluate policyTool, principal & argumentsReady
- Govern executionAllow, require approval, or blockReady
Your policy. The final say.
Run the sample request to see a decision.
This example runs locally. No real tools are called.
ONE GOVERNED PATH
From agent intent
to accountable action.
Custos sits between your agents and the systems they use, applying the same control layer to every request.
AI agentsRequests to act
→CUSTOS CONTROL PLANE
→ Identity Policy Approval Credentials Audit
Enterprise toolsMCP · APIs · services
SHARED CONFIDENCE. DIFFERENT PERSPECTIVES.
Good for your agents.
Great for your people.
FOR THE PEOPLE BUILDING THE FOUNDATIONOne gateway.
One gateway.
Room for every team.
Give teams a consistent way to connect AI to company systems. Manage tools, access, and usage from a shared control plane.
- Organize access by workspace and role
- Review and publish tools in one registry
- Keep usage within configured limits
Your organizationGOVERNED
SSupportCustomer tools
EEngineeringInternal APIs
OOperationsBusiness tools
Shared governance. Scoped access.
FOR THE PEOPLE ASKING THE RIGHT QUESTIONSLess uncertainty.
Less uncertainty.
More accountability.
Define the boundaries of agent access and follow the decisions that matter. Put human review exactly where it belongs.
- Default-deny policies checked on tool calls
- Human approval for sensitive actions
- Audit records with decision context
Identity checked
Policy applied
Decision recorded
ACCOUNTABILITY, BY DESIGNFOR THE PEOPLE MAKING THINGS HAPPENKeep building.
Keep building.
Bring your tools.
Connect MCP servers and APIs through a common gateway. Use scoped execution keys and get a clear outcome for each request.
- MCP, OpenAPI, and generic REST connectors
- One path for governed tool execution
- Correlation IDs for tracing outcomes
governed-tool-call
// Your agent makes a request { "method": "tools/call", "params": { "name": "read_customer", "arguments": { "id": "1042" } } } // Your workspace rules follow it.
TRUST COMES FROM THE DETAILS
Control isn’t an add-on.
It’s the foundation.
Purposeful controls across identity, access,
credentials, and execution.
Identity comes first
SSO sign-in and workspace roles establish who is making a request.
Boundaries stay clear
Tenant-scoped access and default-deny policies govern tool execution.
Credentials stay protected
Encrypted credential storage, write-only secrets, and revocable execution keys.
Usage has limits
Rate limits and plan-based quotas put boundaries on execution and resource growth.
A LITTLE MORE CLARITY
Good questions.
Clear answers.
Getting to know Custos.
What exactly does Custos do?
Custos is a governance gateway between AI agents and your tools. It connects MCP servers and APIs, checks tool calls against your policies, handles approval requirements, and records decisions and outcomes.
Do we need to change our AI models?
No. Custos governs tool access rather than choosing a model for you. Your agent or application needs to send supported MCP or REST tool calls through the Custos gateway. Client transport compatibility should be checked during integration.
Which services can we connect?
Custos includes connectors for HTTP JSON MCP servers, supported OpenAPI 3.0 specifications, and configurable REST APIs. Connections use supported credential types. A large library of native SaaS OAuth integrations is not part of the current product.
How do human approvals work?
A policy can require a separate reviewer before a sensitive tool call may proceed. The approval is bound to the request and can be used once. After approval, the original caller resubmits that request; approving alone does not execute it.
Does Custos replace our security tools?
Custos adds controls around agent tool access. It complements your identity provider, infrastructure security, and operating processes. It is not a prompt-injection scanner, a general data-loss prevention system, or a substitute for a security assessment.
Is the interactive tour connected to a real workspace?
No. The tour uses sample data in your browser to illustrate policy decisions. It does not connect to the gateway, execute tools, transmit your choices, or modify company data.
THE NEXT CHAPTER IS AGENTIC. MAKE IT YOURS.
Let your agents work.
On your terms.
Connect the tools. Set the boundaries. Open up the possibilities.
See Custos in actionSTART A CONVERSATION
Have questions?
Let’s talk.
Talk with the Custos team about agent governance, integrations, or bringing a governed gateway to production.