Runtime compliance for AI agents
KorClaw — The Security Layer for AI Agents
Runtime compliance, policy enforcement, and audit controls for every AI agent action.
Runtime security layer
AI Agent
Requests a sensitive action
KorClaw
Security layer in the execution path
Tools
APIs
Data
Policy decision engine
Every tool call is evaluated before execution. Nothing reaches tools, APIs, or data without a decision.
runtime.trace
agent research-assistant
tool mcp.search.query
action read
status INBOUND
Inbound request from research-assistant…
Security architecture
A security layer between agents and your infrastructure.
KorClaw sits in the execution path — not beside it. Every sensitive action is intercepted, identified, evaluated, and recorded before anything executes.
Layer 01
AI Agents
- Autonomous agents
- Copilots & assistants
- Multi-agent workflows
Layer 02
KorClaw Security Layer
- Runtime interception
- Identity binding
- Policy engine
- Human approval
- Audit evidence
Layer 03
Tools / APIs / Data
- MCP servers
- Internal APIs
- Production data
Execution flow
Core security capabilities
Everything you need to secure agent execution.
Runtime security infrastructure built for production AI agents — from identity to audit.
Runtime security
Every agent action is intercepted in-path before execution. Security is enforced at the moment of the request — not after the fact.
Agent identity
Bind every tool call to a registered agent identity. Know exactly which agent requested what, when, and under which policy scope.
Policy enforcement
Centralized policy engine evaluates identity, tool, action, and risk context. One engine, consistent decisions across every agent.
Tool & API execution control
Govern MCP servers, internal tools, external APIs, and data operations from a single control plane in the execution path.
Human approval
Sensitive operations pause at the runtime until an authorized reviewer approves or denies. No silent execution of high-risk actions.
Audit trail
Immutable evidence for every ALLOW, REQUIRE APPROVAL, and BLOCK. Export-ready records for security, legal, and compliance reviews.
The problem
Why AI agents need runtime control
Once an agent can call a tool, the risk is no longer theoretical. Compliance has to happen in the request path — not in a report after the fact.
Agents act without asking
Autonomous agents call tools, move data, and hit APIs on their own. Post-hoc dashboards cannot stop an action that already executed.
Sensitive actions need a gate
Payments, emails, file writes, and production APIs should never be a direct hop from model output to execution.
Policy must live in the path
Governance only works if it evaluates the request before the tool runs — consistently, for every agent, every call.
Evidence has to be automatic
When security, legal, or customers ask why an agent did something, you need the decision record — not a reconstructed log.
How KorClaw works
A control plane in the execution path.
Every governed action follows the same pipeline: intercept, identify, decide, and record.
AI Agent
An agent requests a tool, MCP action, API, or data operation.
KorClaw Runtime
KorClaw intercepts the call in-path — before anything executes.
Identify & evaluate
The runtime binds agent identity, tool, and action, then applies policy.
Policy decision
KorClaw returns ALLOW, REQUIRE APPROVAL, or BLOCK.
Controlled outcome
Allowed calls proceed. Sensitive ones pause. Blocked ones never reach the tool.
Audit evidence
Every MCP and tool action is recorded as compliance evidence.
Runtime decisions
Allow. Require approval. Block.
One policy engine. Three outcomes. No action reaches MCP, tools, or APIs without a decision.
Proceed to tools
Low-risk, in-policy actions continue to MCP servers, tools, and APIs without delay.
Pause for a human
Sensitive operations hold at the runtime until an authorized reviewer approves or denies them.
Stop before execution
Out-of-policy calls never reach MCP, tools, APIs, or data. The agent receives a governed denial.
Before KorClaw / With KorClaw
Security in the path, not after the fact.
See what changes when agent actions flow through a dedicated security layer.
Tool execution
Agents call tools directly — no gate
Every call intercepted and evaluated in-path
Agent identity
Anonymous or loosely attributed actions
Registered agent identity bound to every request
Policy enforcement
Post-hoc review or manual checks
Real-time ALLOW / APPROVAL / BLOCK at runtime
Sensitive actions
No pause before high-risk execution
Human approval holds before execution
MCP security
Ungoverned MCP tool invocations
Every MCP call checked by policy first
Audit evidence
Reconstructed logs after incidents
Automatic immutable decision records
MCP security
Every MCP tool call is checked by policy before execution.
KorClaw intercepts the path from agent to MCP server so tools never run ungoverned.
MCP runtime path
01
AI Agent
Requests a tool call
02
KorClaw
Intercepts & evaluates
03
MCP Server
Reached only if allowed
04
MCP Tool
Executes after policy
Intercept MCP tool calls
KorClaw sits in front of MCP servers so every tool invocation is captured in-path.
Identify agent and tool
Bind the calling agent to the requested MCP server, tool name, and action before anything runs.
Apply policy before execution
Evaluate identity, scope, and risk against your policies — not after the tool has already acted.
Allow, block, or require approval
Return a runtime decision that either proceeds, holds for a human, or stops the call.
Automatically audit every MCP action
Record the agent, tool, decision, and outcome as immutable evidence for every MCP call.
Audit & compliance evidence
Every decision becomes evidence.
KorClaw records the agent, tool, policy, and outcome automatically — so investigations and audits start from the runtime, not from reconstruction.
- Immutable trail of ALLOW, REQUIRE APPROVAL, and BLOCK
- Agent identity bound to every MCP and tool action
- Export-ready evidence for security and compliance reviews
audit.log
research-assistant · In-policy read against indexed knowledge base
support-agent · Outbound email held for human review
ops-agent · Production payment blocked by spend policy
docs-agent · Repository read permitted for registered agent
KorAI security intelligence
Turn runtime evidence into security intelligence.
KorAI is your compliance copilot — grounded in your audit trails, policies, and controls. Ask questions, surface risks, and get evidence-backed answers.
Compliance posture scoring
Continuous assessment of your agent governance posture with framework-aligned scoring.
Evidence-grounded answers
Ask KorAI about policies, controls, and audit evidence — answers cite only supplied records.
Risk & gap detection
Surface open risks, missing controls, and actionable recommendations from your runtime data.
Security intelligence
Turn audit trails and policy decisions into actionable security insights for your team.
KorAI
You
Which agents triggered approval holds on MCP tools this week?
KorAI
Based on audit evidence: support-agent triggered 3 REQUIRE APPROVAL decisions on mcp.gmail.send. All holds were resolved within policy SLA.
Sources: audit.log · policy.engine · agent.registry
Compliance posture
87/100
Open risks
2
Evidence records
1,247
Developer SDK / CLI / API
Instrument agents. Enforce at runtime.
Evaluate every sensitive action with a few lines of code. Keep your models, tools, and MCP servers — add the security layer in the path.
import { KorClaw } from "@korclaw/sdk"
const kor = new KorClaw(API_KEY)
const decision = await kor.evaluate({
agent: "research-assistant",
tool: "mcp.gmail.send",
action: "send_email",
})
// ALLOW | REQUIRE_APPROVAL | BLOCKBuilt for developers
KorClaw CLI
Install globally, authenticate once, and govern agents from your terminal.
npm install -g @korclaw/clikorclaw loginkorclaw agents listkorclaw policies listkorclaw evaluate --input '{"action":"production_deploy"}'korclaw approvals list --status pendingkorclaw logs --limit 10
Pricing
Runtime governance, priced for production
Start free on any plan. Annual Pro and Business include a 7-day free trial; monthly plans do not.
No free trial on monthly billing
Free
All prices in USD
- 1 Workspace
- Up to 2 AI Agents
- 50 Audit Events/month
- Basic Policies
- Community Support
No credit card required
Pro
No free trial on monthly billing · All prices in USD
- 1 Workspace
- Up to 20 AI Agents
- 1,000 Audit Events/month
- Policy Engine
- Approval Center
- Audit Logs
- SDK & API Access
- MCP & Tool Integrations
- Email Support
No free trial on monthly billing
Business
No free trial on monthly billing · All prices in USD
- Up to 100 AI Agents
- 10,000 Audit Events/month
- Compliance Reports
- Team Management (RBAC)
- Advanced Integrations
- Priority Support
- Scheduled Reports
No free trial on monthly billing
Enterprise
Personalized demo and trial available
- Custom AI Agent Limits
- Custom Audit Event Limits
- Dedicated Support
- SSO & SAML
- Custom SLAs
- On-premise Deployment
- Dedicated Account Manager
Personalized demo and trial available
Usage add-ons
Additional AI Agent
per agent/month
Additional 10,000 Audit Events
/month
Additional Workspace
/month
Additional Admin Seat
/month
Monthly Pro and Business plans do not include a free trial. Switch to Annual to see trial options.
Monthly pricing only. Prices are shown in USD. During checkout, customers will be charged in their local currency using the current live exchange rate.
FAQ
Questions, answered.
Still evaluating runtime control? Here is what teams ask first.
KorClaw is a runtime compliance layer for AI agents. It sits between AI agents and the tools, APIs, and data they use, enforcing policies, approvals, and auditability at runtime.
The security layer for AI agents
Secure every agent action before it executes.
Deploy the runtime security layer your AI agents need. Intercept, identify, decide, and audit — before anything reaches your tools, APIs, or data.