Getting started
From nothing to watching Simian break a real application, check the damage and record exactly what it did — on GKE (about 30 minutes) or on kind on your laptop (about 15).
Documentation
You’re in the simian-agent reference docs. The site root has the marketing pitch; this section is the reference.
Start here
Brand new? → Getting started takes you from an empty project to watching Simian break Online Boutique, check the damage and plan its own faults — on GKE (about 30 minutes) or kind (about 15). Written as a runbook an agent can follow too.
Installing in-cluster? → Deploying with Helm covers the chart install patterns and the recommended overlay.
Picking which chaos engine to use? → Using the chaos engines covers the directed and autonomous patterns for chaos-mesh, network-policy, and envoy-fault — including which one to reach for on which kind of cluster.
Pointing it at GKE? → GKE bring-up covers the Dataplane V2 and Autopilot constraints, and how to measure which fault kinds actually land on your cluster.
Hit something weird? → Known limitations collects the GKE Dataplane V2 NetworkChaos question, the Envoy injection vs gRPC probe interaction, and other gotchas worth knowing about before you debug.
Reference index
Getting things done
- Getting started — GKE or kind, end to end, every step checked.
- Deploying with Helm — in-cluster install patterns.
- Seeing what Simian is doing — live view, what autonomous mode decided and why, the audit trail, metrics and alerts.
- GKE bring-up — the first run against a real GKE cluster, and what lands on Dataplane V2.
- Using the chaos engines — directed and autonomous patterns per engine.
- CLI reference — every flag on every subcommand.
- Helm values reference — every chart value.
Concepts and design
- Design — architecture: Fault Executor chokepoint, LLM Provider contract, chaos drivers, MCP surface, deployment topology.
- Requirements — scope, operating modes, deployment postures, the
R-*requirements catalog. - Roadmap — phased development plan; what’s shipped, what’s next.
- DPv2-compatible chaos engines — why we built
network-policyandenvoy-fault, and what they replace.
Operating
- Known limitations — cluster-side gotchas, dataplane caveats, feature limitations.
Working on Simian
- Contributing — how to file issues, structure PRs, and keep the recommended-defaults overlay honest.
Deploying with Helm
How to install the controller in-cluster via the Helm chart.
Seeing what Simian is doing
Every way to see what Simian has injected, what autonomous mode decided and why, and whether anything was left broken — live, after the fact, and on a dashboard.
The web UI behind Identity-Aware Proxy
Put Simian’s web UI on a real hostname behind Identity-Aware Proxy on GKE, so the people you choose can watch it and inject and clear faults from a browser.
Reading an evaluation: the scorecard
Combine simian-eval runs into one scorecard document and read it in the web UI: scenarios by subjects, what was injected, whether it landed, and what each subject reported. No server needed.
Using the chaos engines
Directed and autonomous patterns for chaos-mesh, network-policy, envoy-fault, and kube-state.
Efficacy probes
Probes gate Apply so a fault that silently did nothing is never reported as applied.
CLI reference
Every flag on every simian subcommand.
Helm values reference
Every Helm chart value, what it does, and the recommended setting.
Known limitations
Cluster-side gotchas, dataplane caveats, and feature limitations contributors should know about.
GKE bring-up
Pointing Simian at a real GKE cluster, and finding out which faults actually land on it.
Architecture at a glance
Three entry points, who launches whom, and which credentials each one holds.
Design
Architecture: Fault Executor chokepoint, LLM Provider contract, chaos drivers, MCP surface.
Requirements
Scope, operating modes, deployment postures, and the R-* requirements catalog.
Roadmap
Phased development roadmap — what’s shipped, what’s next.
Eval substrate plan
Plan for turning Simian into the fault plane and ground-truth source for evaluating the go-steer agentic SRE stack.
DPv2-compatible chaos engines
Why we ship network-policy + envoy-fault engines for GKE Dataplane V2 clusters.
Web UI design
Design doc for a single browser surface serving two purposes: live operator dashboard over simian serve, and the eval scorecard over run artifacts.
Contributing
How to file issues, structure PRs, and keep our chart values overlay honest.