Odigos
Get a demo
Odigos DeepBPF

Deep inside your application. Entirely outside its process.

AI investigations need a sandbox. Odigos DeepBPF captures functions, arguments and return values across Java, Python, Node.js, Go and .NET—from outside the application, through kernel-verified eBPF.

Your application process

JavaPythonNode.jsGo.NET
functionsargumentsreturn values

No instrumentation loaded into the application.

Deep visibility from outside the process

Linux kernel · outside the application

Odigos DeepBPF

eBPF verifier

Checks programs before they are loaded.

eBPF sandbox

Capture executes within kernel-enforced constraints.

Your AI agentRequests scoped capture through Odigos MCP
The verifier checks eBPF programs at load time. Odigos controls the scope of capture requests.

AI chooses what to inspect. Linux verifies the instrumentation.

An AI agent can take a different path through every investigation. Production needs an execution boundary enforced independently of the model. Before an eBPF program is loaded, the Linux verifier checks it against the kernel’s execution rules.

Checked memory access
The verifier tracks pointer types and bounds, rejecting invalid accesses within the eBPF program.
Memory checks in the verifier
Constrained operations
Programs can use only the helpers permitted for their type, with arguments that satisfy the verifier’s checks.
Helper checks in the verifier
Verified before execution
Linux checks the program’s control flow and rejects programs that fail verification before selecting their execution runtime.
Verification in the load path

The depth comes from DeepBPF.

The eBPF sandbox provides the execution boundary. Odigos DeepBPF adds the runtime understanding to see deep into Java, Python, Node.js, Go and .NET entirely from outside their processes.

See the application’s decisions
Connect a request to the functions it called, the arguments they received and the values they returned.
Keep capture outside the application
Collect that context without loading instrumentation into the application’s runtime.
Let the investigation go deeper
Agents request the next capture through MCP. New evidence arrives with subsequent matching traffic in your telemetry backend.
Your team controls the investigation

Set the scope of what AI can inspect.

Choose workloads and capture rules. Apply masking and sampling before telemetry leaves your environment. The Odigos console brings sources, instrumentation, processing and destinations into one view.

The Odigos console: detected sources, in-flight actions and export destinations for a production cluster
Open the full-size console

The verifier checks eBPF program behavior. Your team sets workload access, MCP write permissions and data-handling policies. Those controls work together to govern an AI investigation.

Configure agent access
The Production Context Platform

Stop guessing. Ask production.

One agent. One service. See how Odigos helps answer a production question.