Documentation

AgentOps is a Kubernetes-native platform for deploying and operating AI agents as first-class workloads. These docs cover everything from installing the operator to writing custom tool servers.

Getting Started

Install the AgentOps operator on your cluster, deploy your first agent, and verify it’s running. Covers prerequisites, Helm chart installation, and a minimal Agent CR walkthrough.

Concepts

How things work under the hood:

• Agent Lifecycle – how the operator reconciles Agent CRs into running pods with sidecars, storage, and networking.
• Three-Layer Memory – working memory, short-term session summaries, and long-term observations. BM25 relevance ranking, three-tier write dedup, deterministic summarization.
• Delegation – parallel sub-agent spawning via Kubernetes Jobs, structured result aggregation, concurrency control.
• MCP Tools – tool servers as OCI artifacts, stdio transport, registry resolution, and sidecar injection.
• Channels – Slack, webhook, and custom channel bridges for agent I/O.
• FEP Streaming – the Fantasy Event Protocol over SSE, how the console connects to live agent sessions.

Guides

Task-oriented walkthroughs:

• Writing a custom MCP tool server
• Configuring agent memory for your use case
• Setting up agent delegation chains
• Connecting agents to Slack channels
• Deploying the AgentOps Console
• Packaging and pushing OCI tool artifacts

Reference

API and specification details:

• CRD Reference – full spec for Agent, Tool, Channel, and Resource custom resources.
• Runtime Configuration – environment variables, flags, and runtime behavior.
• Memory API – REST endpoints for the agentops-memory service.
• FEP Protocol – event types, payloads, and SSE stream format.
• Helm Values – all configurable values for the agentops-platform chart.

Project

Contribution guidelines, release process, architecture decisions, and roadmap. Includes the planned OCI Skills feature (spec.skillRefs) for injecting static expertise as markdown artifacts into agent system prompts.