Run it. Read a finding. Approve a fix.
Agentshield maps how your Kubernetes and AWS workloads really talk, investigates what changes with evidence, and proposes fixes you approve. These pages cover running it, how it reasons, and how to extend it.
Reading order for newcomers
- What AgentShield does: the problem, the pipeline, and the components.
- Getting started: run it locally.
- Glossary: the vocabulary used everywhere else. If you read only one part, read the five terms you need at the top.
- Data model: identities, assertions, and contributions.
- Architecture overview: what runs where.
Then pick the area you are working on.
Concepts
Architecture
- Overview: process layout, data flow, HTTP API, known limits
- Publication and storage: contributions, generations, PostgreSQL, change feed, migrations
- Change feed and reasoning: detectors, model investigation, validation, findings
- Telemetry: ingestion, event schema, rollups, backpressure
- Detection rules: CEL rules and model-authored Deployment rules
- Remediation: suggestions, actions, approval, verification
Plugins
- Plugin contract
- Tutorial: the life of a plugin
- Style guide
- Providers: Kubernetes, Istio, Cilium, agentgateway, AWS
Operations
- Configuration: every flag and environment variable
- Helm deployment: installation, PostgreSQL, ClickHouse, and access controls
- Security model
Development
Browse the docs
Getting started
Try the demo, run agentshield against Kubernetes or AWS, open the explorer, and turn on the model features.
→Concepts
How plugins, the graph, telemetry, detection, reasoning, and remediation fit together.
→Plugins
The plugins in the box, what each reads, and what is planned.
→Operations
Configure and deploy the service, understand its security model, call its API, and measure it with evals.
→Architecture
Process layout, data flow, plugin wiring, HTTP endpoints, and failure behavior.
→Contributing
Build and test the Go service and the web UI, and find where each part lives.
→Development
Testing and developing agentshield.
→