The first cloud-native agent harness
Own the harness, rent the model
What can you do with Stacklok’s agent harness?
We’re so glad you asked! You can:
- Stop trying to lift and shift desktop harnesses to the cloud
- Build your own harness from our library of components
- Leverage your existing software infrastructure; no new overhead to operate
- Scale fleets of agents in the cloud with Kubernetes patterns
- Retain ownership of your context, tools, policies and more, so you can move between models and providers
“Mecatl takes the familiar harness, blows it up, and puts it back together to be more scalable, more secure, and able to serve more kinds of clients. We are designing something that works with cloud systems so that hosting, exposure and security of these more capable agent systems falls out of the harness architecture.”
What makes Stacklok’s agent harness different?
Cloud-native design
Alternatives: Assumes use by one developer, on one machine with one local filesystem.
Stacklok: Separates sessions, memory, tools, sandboxes and more into extensible interfaces that can be deployed and scaled independently
Agnostic
Alternatives: Agent harnesses (e.g. Claude Code, Codex, etc.) bind the harness and the model, so you’re locked into the provider
Stacklok: Ensures you own the harness, so you can rent the model, moving between providers to save on AI costs and get maximum efficiency
Secure
Alternatives: Most harnesses couple together the agent loop and the sandbox. Meaning execution and untrusted actions share one process.
Stacklok: By separating the agent loop from the sandbox, we keep bad code outside your trust boundary.
Composable
Alternatives: Most harnesses are opinionated. Most frameworks force you to bring-your-own-everything.
Stacklok: We supply the agent loop, and you connect your tools, policies, etc. through explicit interfaces.
Fun fact:
Stacklok’s Kubernetes operator enables 20M+ tool calls every month.
Want to talk through your options?