Get your harness off the desktop
Start locally, then deploy a fleet of agents on Kubernetes with your choice of cloud and model provider.
We blew up the harness and then put it back together: more secure, more scalable, more extensible.

Built from the ground up to run in the cloud. Get your harness off your laptop and run fleets of agents anywhere.
Start locally, then deploy a fleet of agents on Kubernetes with your choice of cloud and model provider.
Separate control from execution to keep bad code outside your trust boundary.
Use Mecatl's library of components to assemble a custom harness.
Set permissions and choose which tool calls require your approval.

Mecatl gives you the hard parts. Use what you need.
Multi-turn reasoning, tool dispatch, compaction, permission gating, and event streaming are built in.
Modular design. Swap LLM backends, persistence, distributed locks, or permission logic without touching the engine. Reference adapters included for every port.
Disposable process, externalized state, durable event record. mecak8s pre-wires Redis and Kubernetes leases so you can scale stateless pods from day one.
github.com/stacklok/mecatl/engine is its own Go module with a tiny dependency closure. Embed the loop directly, no server required.
OpenTelemetry traces, LLM resilience decorator, slow-turn ring buffer, and an MCP-accessible perf server.
Every agent acts under a verifiable identity. Mecatl stamps identity at every agent action, not just at the login gate.
mecatui, or connect to mecated over gRPC or HTTP/SSE.Cloud-nativemecak8s: stateless pods, Redis-backed state, and multi-replica coordination.CIUse mecatequi to turn one prompt into a patch and a pass/fail result for your pipeline.EmbedImport the engine directly into your Go project.Featured project
The open source MCP platform trusted by enterprises, securing 10M tool calls each month.