Kagent — Kubernetes Agent Framework (Solo.io)

Activities: Y/M	Diagnostics: Y/H	Provisioning: Y/M
Event ontology: P/M	Observability: Y/H	Confidence: High

Build style / interface — Kubernetes-native, open-source agent framework deployed in-cluster using CRDs and controllers; optional UI and CLI for agent management.
What it actually does — Runs AI-powered agents inside Kubernetes clusters to diagnose issues, inspect live state, execute remediation actions, and orchestrate operational workflows using tools such as kubectl, Helm, Argo, and Prometheus queries.
Data / telemetry — Live Kubernetes API objects, cluster events, pod logs, and metrics retrieved on demand from Prometheus/Grafana and other in-cluster observability tools.
Interoperability — Strong cloud-native integration via Kubernetes APIs; supports MCP (Model Context Protocol) for tool access and A2A patterns for agent-to-agent communication; integrates with Prometheus, Grafana, Istio, Argo CD, Helm, and external LLM providers.
Deployment model — Self-hosted, runs entirely inside Kubernetes clusters; open-source (CNCF Sandbox project).
Notes — Focused on Kubernetes and platform engineering use cases; governance, approvals, and RBAC rely on Kubernetes-native controls and external policy engines rather than a built-in ITSM layer.

---

🧠 UKM Snapshots

ingest Y/M, index P/M, retrieval Y/M, governance P/L, overall medium

Note: Knowledge derived from live cluster state, configuration, and retrieved documentation; no centralized long-term enterprise knowledge base by default.

---

⚙️ UAM Snapshots

ingest Y/M, index Y/M, retrieval Y/M, governance P/M, overall medium-high

Note: Captures agent actions, tool invocations, and remediation steps as structured activity flows; replay and approval depend on surrounding platform controls.

---

🔍 UDM Snapshots

ingest Y/H, index Y/M, retrieval Y/M, governance Y/M, overall high

Note: Strong Kubernetes-centric diagnostics using live queries, logs, events, and dependency inspection; RCA is evidence-driven but scoped to cluster context.

---

👁️ UOM Snapshots

ingest Y/H, index Y/M, retrieval Y/H, governance Y/M, overall high

Note: Leverages existing observability stacks (Prometheus, Grafana, logs) with deep contextual navigation rather than storing telemetry itself.

---

🧬 UEOM Snapshots

ingest P/M, index P/M, retrieval P/M, governance P/M, overall medium

Note: Entity model centers on Kubernetes primitives (clusters, namespaces, pods, services, workloads) and their relationships.

---

🚀 UPM Snapshots

ingest Y/M, index Y/M, retrieval Y/M, governance P/M, overall medium-high

Note: Supports controlled execution of kubectl, Helm, and Argo-based changes; rollback and safety depend on underlying Kubernetes and GitOps tooling.

---

Latest updates — Accepted into CNCF Sandbox; rapid community growth and expanded MCP-based tool integrations (2024–2025).
Links —

• https://kagent.dev
• https://docs.kagent.dev
• https://github.com/kagent-dev/kagent