Agent↓Tool↓System
Direct access makes permissions and outcomes difficult to control.
USE CASES
Put AI agents to work —
without unchecked access.
Custos sits between AI agents and enterprise tools, enforcing identity, policies, approvals, controlled execution, and auditability.
AI agentIntent to act
CUSTOS CONTROL PLANE
IdentityPolicyApprovalCredentialsAudit
Enterprise toolsMCP · APIs · internal services
THE CORE IDEA
AI can act.
Custos sets the boundaries.
A governed path turns agent intent into an explainable tool decision.
Agent↓CustosIdentity · Policy · Approval↓ALLOW / APPROVAL / DENY↓Tool + audit
Every request is evaluated before it reaches the upstream tool.
GOVERNED ACTIONS
Whatever the agent needs to do,
your policies stay in the path.
Custos governs actions through connected APIs, MCP servers, and internal tools. It is a control layer—not a native integration catalog.
Agent wants to…Custos can govern it through a connected tool
📧 Send an emailConnected API / MCP tool
🎫 Create a Jira ticketConnected Jira API / MCP tool
👤 Update SalesforceConnected Salesforce API / MCP tool
💰 Issue a refundApproval policy before execution
🧾 Create a purchase orderConnected procurement API
👥 Access customer dataGoverned customer-data tool
🗄️ Modify a databaseConnected DB / API tool
🚀 Deploy softwareConnected deployment API
🔄 Restart infrastructureConnected infrastructure API
☁️ Create cloud resourcesConnected cloud API
🗑️ Delete recordsDeny or require approval
🔌 Call an internal APIDirectly aligned with Custos
WHERE CUSTOS FITS
Useful wherever
AI needs to act.
Use the same governance layer across financial, customer, engineering, and internal business workflows.
01
FINANCIAL OPERATIONS
Let AI handle financial actions safely.
Govern refunds, billing actions, and financial APIs while keeping high-value decisions with an authorized reviewer.
AGENT REQUEST“Refund $10,000 for Order #123.”HIGH RISK
- 01Request received
- 02Identity and arguments verified
- 03High-value action requires approval
- 04Reviewer approves; action executes and is recorded
High-value actions→Human approval→Controlled execution→Audit
02
CUSTOMER SUPPORT
Give AI controlled access to customer systems.
Agents can retrieve orders, update records, or issue refunds without receiving unrestricted access.
Read orderALLOW
Update customer addressALLOW
Issue high-value refundREQUIRE APPROVAL
Delete customer accountDENY
Access is controlled at the tool and action level.
03
ENGINEERING & DEVOPS
Let AI operate infrastructure with guardrails.
Govern access to internal infrastructure APIs while sensitive operational actions remain controlled.
Restart a serviceQuery deployment statusInspect logsTrigger an operational workflow
Example policy AI can restart development services automatically, but production changes require approval.
04
ENTERPRISE APIS & INTERNAL TOOLS
Connect AI agents to the tools your business already uses.
Custos provides one governed path between agents and MCP servers, APIs, and internal tools.
MCP serversCRM APIsOrder systemsFinance APIsSupport systemsInternal developer tools
No native SaaS integration is required: Custos governs supported tool calls at runtime.
SENSITIVE OPERATIONS
Not every agent should
perform every action.
AGENT A
Read customer data ALLOW
Modify customer data REQUIRE APPROVAL
AGENT B
Administrative operation DENY
Every decision follows policy.
AI GOVERNANCE & AUDIT
Know what your AI agents
are doing.
Every important action should be explainable from request to result.
- Request received
- Identity verified
- Tool selected
- Policy evaluated
- Approval requested
- Approved
- Tool executed
- Result recorded
AUDIT EVENT
Agentsupport-agent
Toolissue_refund
Arguments$10,000 / Order #123
PolicyHigh Value Refund
DecisionREQUIRE_APPROVAL
Approved byfinance-reviewer
ResultSUCCESS
WHAT CUSTOS CONTROLS
The controls that keep
agent actions accountable.
Identity
Know which user, agent, or client is making the request.
Tool access
Control which tools agents can discover and use.
Policies
Allow, deny, or require approval based on the action.
Human approval
Route risky actions to an authorized reviewer.
Credentials
Protect upstream credentials from direct agent exposure.
Execution & audit
Validate governed requests and record important outcomes.
HOW CUSTOS WORKS
Autonomous where safe.
Controlled where risky.
AI agent
↓CUSTOSIdentity → Tool & argument validation → Policy
↓ALLOWREQUIRE APPROVALDENY
↓Enterprise tool / API
↓Audit
WHO USES CUSTOS
One control layer.
Different teams.
One governed path for AI integrations.
Policy, approvals, and audit around agent actions.
Give AI access to tools without rebuilding every integration.
COMMON QUESTIONS
Good questions.
Clear answers.
What is Custos?
Custos is the control plane between AI agents and the tools they use. It governs supported tool calls with identity, policy, approval, controlled execution, and audit.
Where does Custos sit in an AI architecture?
Agents send supported MCP or API tool calls through Custos before they reach enterprise tools, internal services, or upstream APIs.
Does Custos replace IAM?
No. IAM establishes identity and access controls. Custos governs what an AI agent is allowed to do through connected tools at runtime.
How does human approval work?
A policy can require a separate reviewer before a sensitive tool call proceeds. Approval is bound to the request and can be used once.
Does Custos work with MCP and APIs?
Custos supports HTTP JSON MCP servers, supported OpenAPI specifications, and configurable REST APIs. Compatibility should be checked during integration.
What gets audited?
Custos records request context, policy decisions, approval outcomes, and execution results so important actions can be understood later.
CONTROL THE NEXT ACTION
Give your agents access.
Keep control.
Connect your tools, define your policies, and let AI operate within the boundaries your organization defines.