Awesome DevOps Agents

A curated, evidence-oriented map of AI agents for DevOps, SRE, AIOps, incident response, and root-cause analysis (RCA).

This list is for engineers deciding what to build or evaluate, researchers tracking the field, and maintainers looking for reproducible systems—not just another collection of AI links. Product capabilities are mapped to the incident lifecycle, while evidence and maturity are recorded separately.

Last reviewed: 2026-07-31 · Catalog version: 0.2.0

• Start here
• Scope and labels
• Field signals
• Open-source systems
• Commercial products
• Benchmarks and environments
• Research papers
• Tools, protocols, and datasets
• Foundations
• Model capability and economics
• Related awesome lists
• Contributing
• License and disclaimer

Included resources materially support at least one stage of the operational loop:

detect → triage → localize → diagnose → plan → mitigate → verify → learn

The detailed definitions are in docs/taxonomy.md. Short labels used here:

• Maturity: GA, Preview, Active OSS, Research, or Early.
• Evidence: E3 reproducible external evaluation; E2 published methodology or artifacts; E1 technical documentation/demo; E0 claim only. Evidence is not product quality.
• Autonomy: L0 assist only; L1 recommend; L2 execute with approval; L3 bounded automatic execution; L4 broad autonomy. See safety guidance.
• Company/steward: the organization currently publishing or stewarding the resource; it does not imply sole authorship.

Commercial capabilities below are vendor-reported unless an independent evaluation is linked. Inclusion is not endorsement.

The fast-changing part of this repository is summarized in docs/field-signals.md. The current signals are:

• evaluation is moving from static answers toward live, stateful environments and complete trajectories;
• evidence acquisition and inter-agent communication remain major failure modes, not merely model-size problems;
• causal/topology grounding is being measured as a way to reduce investigation search and tool cost;
• open-source systems increasingly combine background alert ingestion, parallel investigation, memory, and post-RCA workflows;
• project status changes quickly—archived systems and vendor lifecycle changes must be tracked, not silently forgotten.

The machine-generated catalog report exposes coverage by type, lifecycle, evidence level, maturity, and common tags.

This compact view names the actual product and company. Status and access can vary by region, plan, and date; follow the official link for current availability. A fuller feature/evidence matrix is in docs/product-landscape.md.

• OpenTelemetry Demo — instrumented microservice system with feature flags and fault scenarios.
• GoogleCloudPlatform/microservices-demo — Online Boutique cloud-native reference application.
• DeathStarBench — microservice benchmark suite with realistic service graphs.
• Chaos Mesh and LitmusChaos — controlled fault injection for generating evaluation episodes.

• RCACopilot: Web Service Root Cause Analysis via Large Language Models (2023) — early industrial design combining diagnostic information collection, ranking, and report generation.
• RCAgent: Cloud Root Cause Analysis by Autonomous Agents with Tool-Augmented LLMs (2023/2024) — tool-augmented agent loop and one of the central references for LLM-based cloud RCA. ACM version.
• Exploring LLM-based Agents for Root Cause Analysis (2024) — studies agent design and observability-tool interaction for RCA.
• Cloud Atlas: Efficient Fault Localization for Cloud Systems using LLMs (2024) — combines LLM reasoning with cloud-system context for localization.
• AIOpsLab: A Holistic Framework to Evaluate AI Agents for Enabling Autonomous Clouds (2025) — agent–environment interface, workload orchestration, fault injection, and evaluation.
• ITBench: Evaluating AI Agents across Diverse Real-World IT Automation Tasks (2025) — end-to-end SRE, CISO, and FinOps scenarios with sobering baseline resolution rates.
• Flow-of-Action: SOP Enhanced LLM-Based Multi-Agent System for Root Cause Analysis (2025) — injects standard operating procedures into a multi-agent workflow.
• OpenRCA: Can Large Language Models Locate the Root Cause of Software Incidents? (2025) — a large, multimodal, real-system RCA benchmark and baseline study.
• RCAFlow: Retrieval-Augmented Generation for Root Cause Analysis (2026) — connects retrieval and structured RCA workflows.
• Stalled, Biased, and Confused: Failure Modes of LLM Agents in Cloud Incident Diagnosis (2026) — taxonomy and analysis of investigation failures.
• Why Do AI Agents Systematically Fail at Cloud Root Cause Analysis? (2026) — highlights partial observability, evidence acquisition, and stopping/commitment failures.
• Cloud-OpsBench: Benchmarking AI Agents for Cloud Operations (2026) — scalable environment snapshots and broad incident coverage.
• LATS-RCA: Language Agent Tree Search for Root Cause Analysis (2026) — explores explicit search over diagnostic trajectories.
• Towards In-Depth Root Cause Localization with Agentic Reasoning (2026) — moves from coarse service localization toward deeper causal explanation.
• Causely: A Causal Intelligence Layer for Enterprise AI (2026) — controlled study of how causal/topology grounding changes diagnosis accuracy, latency, tool calls, tokens, and cost; results are first-party and should be independently replicated.
• Large Language Models for Agentic NetOps and AIOps (2026) — survey organized around autonomy, evidence traces, assurance contracts, workflow evaluation, canaries, and rollback-aware safety.

• ReAct: Synergizing Reasoning and Acting in Language Models (2022) — the basic reasoning/action/observation loop behind many operational agents.
• Reflexion: Language Agents with Verbal Reinforcement Learning (2023) — reflection and episodic feedback without parameter updates.
• Automated Design of Agentic Systems (2024) — searches over executable agent designs rather than hand-selecting one fixed workflow.
• AFlow: Automating Agentic Workflow Generation (2024) — optimizes workflows represented as code through iterative search.
• Multi-agent Architecture Search via Agentic Supernet (2025) — treats candidate multi-agent architectures as paths in a supernet and adapts architecture to tasks.
• Towards a Science of Scaling Agent Systems (2025) — empirical laws for when additional agents help or hurt, emphasizing task properties and coordination overhead.

• MicroRCA: Root Cause Localization of Performance Issues in Microservices (2020) — topology-aware anomaly propagation baseline.
• PyRCA: A Library for Metric-based Root Cause Analysis (2023) — practical library and benchmark of causal/graph-based methods.
• A Goal-Driven Survey of Root Cause Analysis (2025) — organizes RCA by goal, signals, assumptions, and evaluation rather than by model name.
• RCAPapers — a broader paper index when this intentionally selective list is not enough.

• Model Context Protocol — standard interface for connecting models to tools and contextual data.
• Agent2Agent Protocol — interoperability protocol for communicating agents.
• OpenTelemetry — vendor-neutral traces, metrics, and logs; the natural observation layer for portable agents.
• Prometheus HTTP API and Grafana HTTP API — foundational structured observability interfaces.

• OpenRCA data — multimodal failure data and benchmark tooling.
• Loghub — public log datasets for log parsing and anomaly research; not an end-to-end agent benchmark.
• AIOps Challenge datasets — historical anomaly/RCA data used by AIOps research.

• Site Reliability Engineering — focus on monitoring distributed systems, effective troubleshooting, incident management, automation, and postmortems.
• The Site Reliability Workbook — practical SLOs, alerting, incident response, and operational implementation.
• Google Cloud Architecture Framework: Operational Excellence — operational readiness and lifecycle framing.
• Incident Command System — useful background for roles, escalation, and coordination during high-severity incidents.

• Anthropic: Building Effective Agents — practical distinction between workflows and agents, with compositional patterns.
• LangGraph, AutoGen, Semantic Kernel, and CrewAI — general orchestration frameworks; inclusion here does not imply DevOps-specific quality.
• OpenAI Agents SDK — agent primitives, handoffs, guardrails, sessions, and tracing.

• NIST AI Risk Management Framework — general AI governance and risk vocabulary.
• OWASP Top 10 for LLM Applications — prompt injection, excessive agency, sensitive information, and supply-chain risks.
• SLSA — software supply-chain integrity relevant to agent-generated changes and tools.

General model rankings are supporting signals, not substitutes for an RCA benchmark.

• Artificial Analysis Intelligence Index — composite capability index; always record its version and date.
• Artificial Analysis methodology — explains benchmark composition and weighting.
• Artificial Analysis API — useful for a versioned model-cost/latency dashboard.
• SWE-bench — software engineering capability; relevant to repair, less so to observation and diagnosis.
• LiveBench — contamination-aware general model evaluation.
• HELM — transparent, multi-metric evaluation framework.

For operational agents, track at least: task success, root-cause precision and ranking, evidence quality, calibration/abstention, time and tool calls, cost, latency, recovery from misleading observations, safety violations, and human effort. See docs/evaluation.md.

• awesome-devops-ai — broad AI × DevOps tools and resources.
• awesome-ai-sre — AI-assisted SRE resources.
• awesome-sre-agents — SRE agent ecosystem.
• awesome-LLM-AIOps — research-heavy LLM-for-AIOps collection.
• awesome-harness-engineering — engineering patterns around agent harnesses.
• awesome-sre — broad SRE foundations.

Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request. Every addition must state:

1. the operational stage it supports;
2. who publishes or maintains it;
3. maturity and evidence level;
4. why it is distinct from existing entries;
5. an official or primary source—not only a launch tweet or aggregator.

The companion machine-readable index is catalog/resources.yaml, validated against catalog/schema.json.

To the extent possible under law, this curation is released under CC0 1.0. Linked projects, papers, names, and trademarks retain their own licenses and rights.

This list is informational. Validate security boundaries, data handling, permissions, and rollback behavior before allowing any agent to operate production systems.