Runtime Intelligence

The evidence layer that reads what actually runs.

Through memory analysis, Kodem reads running workloads to determine what code and AI capabilities are loaded, what executes, and which paths are reachable. That evidence grounds application risk, reconstructs agent architecture, and verifies remediation.

US Patent 11,989,572 B2

US Patent 12,645,568 B2

One foundation. Every security outcome.

The same runtime evidence grounds application security, runtime AI security, and verified remediation.

Application security

Loaded, reachable, exploitable

Is the vulnerable code loaded into memory?

Does a path reach it from real input?

Did the function actually execute?

What it determines: Which vulnerable code is loaded, reachable, and relevant in the running environment.

Runtime AI security

Agent-aware intelligence

Which agents, models, and tools are loaded?

What can each agent reach?

Where can data flow, and through which delegation?

What it determines: Loaded agents, models, tools, and delegation paths, attributed to their runtime origin.

Verified outcomes

Proof.
Not assumptions

Is the risky path still reachable after a fix?

Did the change hold in production?

Which risks can be safely deprioritized?

What it determines: Whether remediation removed the risky path, re-checked against runtime state.

The evidence model

Most tools stop at present. Runtime intelligence measures how real a finding is, across distinct evidence levels. They are not always a strict linear sequence.

Present
The code or package exists in the application artifact.
Loaded
A running process loaded it into memory.
Reachable
A path exists from an exposed input or application flow to the vulnerable code.
Executed
The relevant function was observed running. This does not by itself mean it was exploited.
Exploitable
The runtime conditions required for exploitation are present, supported by exploit evidence or controlled validation.

How Runtime Intelligence works

Runtime intelligence is built on memory analysis. Kodem inspects the memory of running processes to reveal the code and capabilities loaded inside them, then correlates that state with execution and operating-system evidence to determine which functions run and how the application behaves in its deployed environment.

1

Read what is in memory

User-space memory analysis reads each running process to reveal the code and AI components actually loaded inside it, with kernel-level signals capturing load and execution events.

2

Correlate the evidence

Kodem joins memory state with execution, operating-system, and code evidence to see what is reachable and what runs.

3

Grade what's real

Each finding maps to the evidence levels above, so triage starts from what actually runs, not what is theoretically present.

4

Verify in runtime

After a change, runtime evidence shows whether the risky path still loads and executes, so a fix is confirmed against what runs, not assumed.

Deployed the way your environment allows

Because memory is the primary evidence source, Kodem is not limited to a single kernel mechanism. Runtime intelligence supports Linux and Windows and can operate through continuous in-host collection or sensorless analysis in restricted and air-gapped environments.

Runtime intelligence operates inside your environment. Where connectivity is used, communication is outbound only. Kodem analyzes memory inside the customer environment and sends only the metadata required to generate runtime intelligence. Source code, secrets, and raw memory contents are not transmitted.

In-host sensor

Continuous collection

Runs alongside the workload, not inside the application process. Continuous, deep runtime intelligence across Linux and Windows, deployed in about five minutes.

Sensorless analysis

For restricted environments

Analyzes what is running from process memory captured out of band, with no in-process sensor and no external connection. It operates below the OS for tamper resistance and runs fully air-gapped.

Cloud environments, on-premise environments, air-gapped environments, and highly regulated environments.

One runtime foundation for code, dependencies, and AI agents

Because runtime intelligence reads what is loaded in the running workload, the same foundation reveals both the software and the AI layer.

RBOM

Runtime Bill of Materials

Your inventory enriched with runtime evidence, reflecting what actually loads and executes. It surfaces shadow dependencies that load at runtime but never appear in your manifest.

Prioritization

Kodem Score

A per-issue score that combines severity and exploit signals with runtime factors like whether the issue is in runtime, internet-exposed, and on an ingress application.

Agent-aware

The AI layer

Reconstructs agents, models, tools, MCP servers, guardrails, and their delegation from runtime evidence, with reachable AI risk mapped to the OWASP LLM Top 10. Discovery and attribution ship today; enforcement is the direction.

Frequently Asked Questions

What is runtime intelligence?

Runtime intelligence is continuous visibility into how an application behaves in production: which code and capabilities load, which functions execute, and which paths are reachable.

Kodem builds it from runtime evidence through memory analysis, and it is the patented foundation the rest of the platform runs on.

It removes the large share of findings that never load or run.

How is it different from static analysis?

Static analysis describes what could happen from the code.

Runtime intelligence establishes what is real by reading the running process in memory and correlating it with execution and operating-system evidence, so findings that never load or run drop out.

Does runtime intelligence cover AI applications?

Yes. The same foundation reconstructs the AI layer from runtime evidence: agents, sub-agents, models, tools, MCP servers, guardrails, and their delegation relationships, including loaded components that have not yet generated traffic. Reachable AI risk maps to the OWASP LLM Top 10.

  • Discovery and attribution ship today.
  • Agent-aware enforcement is the direction Kodem is building.
Does it require code changes or instrumentation?

No. Kodem uses out-of-band memory analysis, and the in-host sensor runs alongside the workload, not inside your application process.

It does not modify code or add in-process instrumentation, and for restricted or air-gapped environments Kodem can run sensorless.

Does Kodem use eBPF?

Kodem's core evidence source is memory analysis, not a single kernel mechanism. It can use kernel-level techniques where useful, but because memory analysis is the foundation, runtime intelligence is not limited to the Linux kernel: it supports Windows, reaches endpoints, and can run sensorless in air-gapped environments.

Which languages have function-level runtime visibility?

Function-level runtime execution is supported by language, including Python, Go, Java, JavaScript, Kotlin, Ruby, Scala, and TypeScript. Full language and package-manager coverage is documented separately.

What data leaves my environment?

Only the metadata required to generate runtime intelligence: package and function names, file paths, execution status, and component metadata. Source code, secrets, and raw memory contents are not transmitted.
Self-hosted solutions, do not require any data to leave your environment.

See Runtime Intelligence in your own environment

Discover which code and AI capabilities are loaded, which paths are reachable, and which risks require action.

See it on your applications
See it on your applications
Illustration of a nighttime garden landscape with an infinity-shaped lake under a crescent moon