The best AIOps and AI SRE tools in 2026

Updated November 15, 2026By HostCrust11 tools compared11 min read

HolmesGPT is the best open-source AI SRE: a CNCF sandbox agent that investigates alerts by querying Prometheus, Loki, Grafana, Datadog and Kubernetes, read-only by design, with the LLM of your choice. Datadog Bits Investigation is the best if your telemetry already lives in Datadog, with unusually clear pricing of about 4 AI credits per investigation. PagerDuty AIOps is the best for cutting alert noise across many monitoring tools.

Every tool here can be wrong. Prefer ones that show their evidence and need a human to approve any change.

Our top picks

Compare at a glance

ToolBest forPricingOpen sourceDeploymentTested by us
HolmesGPTa free, open-source AI investigator you controlFree (you pay your LLM provider)YesSelf-hosted CLI, server or operatorDocs only
Datadog Bits Investigationteams whose telemetry already lives in DatadogBilled in AI credits: $500 per 500 credits per month billed annually, or $1.30 per credit on demand; an autonomous investigation is estimated at about 4 creditsNoSaaS (part of Datadog)Docs only
PagerDuty AIOpscutting alert noise across many monitoring toolsAIOps add-on: no public price; Incident Management Professional $21 per user per month billed annually ($25 monthly); PD Reliability Platform from $2,800 per year (Starter), with Signal Intelligence included from Essential ($12,000 per year)NoSaaSDocs only
incident.io AI SREAI investigation inside your incident processBasic free; AI agent included on Pro ($25 per user per month) and Enterprise; root cause analysis and the autonomous agent are add-ons with no public priceNoSaaSDocs only
Clerica standalone AI SRE with published pricingTeam $700, Pro $1,500 and Scale $3,000 per month; credits cost $0.50 ($1 per change verification, $5 per issue investigation); 1,000 trial creditsNoSaaS, with a connector for private resourcesDocs only
Resolve AIlarge teams that want an agent on primary on-callNo public price (contact sales)NoSaaSDocs only
Rootly AI SREteams that already run incidents in RootlyAI SRE: contact sales (requires Incident Response at $20 per user per month)NoSaaSDocs only
Dynatraceenterprises that want AI inside full-stack observabilityFull-Stack Monitoring $58 per month per 8 GiB host; Infrastructure Monitoring $29 per host per month; logs $0.20 per GiB ingested; AI exploration included; Agentic Operations $0.10 per AI unitNoSaaS with an agent on each hostDocs only
BigPandalarge IT operations teams drowning in alertsNo public price (contact sales)NoSaaSDocs only
KomodorKubernetes-heavy teamsNo public price: custom pricing made of a platform fee plus AI tokens; free trialNoSaaS, with optional self-hosted agents; a self-hosted control plane by agreement (beta)Docs only
K8sGPTa free first look at what's wrong in a Kubernetes clusterFree to use under its open-source license (Apache-2.0); you pay for any hosted AI backendYesSelf-hosted CLI or in-cluster operatorDocs only

The full ranking

HolmesGPT

Best for: a free, open-source AI investigator you control

HolmesGPT is the strongest open-source option. It's an agent that investigates alerts by querying live data from Prometheus, Grafana, Loki, Datadog, Kubernetes and custom REST APIs, and it can pull alerts from Alertmanager (and, from its CLI, PagerDuty, OpsGenie and Jira) and write findings back. It doesn't require Kubernetes, works with OpenAI, Anthropic, Azure, Bedrock, Gemini and other model providers, and is read-only by design, respecting RBAC; optional toolsets can scale, roll back or open pull requests. It's a CNCF sandbox project, created by Robusta and licensed Apache-2.0. You supply and pay for the model, and you own the setup and tuning.

Pricing
Free (you pay your LLM provider) (checked November 15, 2026, source)
License
Open source (Apache-2.0)
Deployment
Self-hosted CLI, server or operator
Runs on
Kubernetes, VMs, cloud providers and SaaS tools; bring your own LLM (OpenAI, Anthropic, Azure, Bedrock, Gemini and others)

Pros

  • Free and open source (Apache-2.0)
  • Read-only by design; write toolsets are optional
  • Works with your existing Prometheus, Loki, Grafana or Datadog
  • Bring your own model provider

Cons

  • You run and tune it yourself
  • Model costs and what data reaches the model are yours to manage
  • Sandbox-stage CNCF project

Not tested by us yet: based on the vendor's documentation.

Datadog Bits Investigation

Best for: teams whose telemetry already lives in Datadog

Datadog's always-on SRE agent investigates each alert as it fires, explores several possible root causes in parallel, and returns an analysis you can verify, with Slack, Jira, ServiceNow and GitHub integrations. Its pricing is unusually clear for this category: AI credits cost $500 per 500 a month billed annually, or $1.30 each on demand, and an autonomous investigation is estimated at about 4 credits, so roughly $4 per investigation at the annual rate by our math. It only helps if your metrics, logs and traces are already in Datadog.

Pricing
Billed in AI credits: $500 per 500 credits per month billed annually, or $1.30 per credit on demand; an autonomous investigation is estimated at about 4 credits (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS (part of Datadog)
Runs on
Datadog customers; investigates using data already in Datadog, plus Slack, Jira, ServiceNow and GitHub

Pros

  • Investigates alerts automatically as they fire
  • Published per-credit pricing
  • Uses the data you already send to Datadog

Cons

  • Only for Datadog customers
  • Credit use varies with task complexity
  • Proprietary

Not tested by us yet: based on the vendor's documentation.

PagerDuty AIOps

Best for: cutting alert noise across many monitoring tools

PagerDuty AIOps, called Signal Intelligence on its newer platform plans, works on the stream of events from your monitoring tools: alert grouping, suppression and thresholds, event orchestration, and root-cause hints from past incidents, related incidents, change correlation and Probable Origins. PagerDuty claims it can cut alert noise by 91%. On Incident Management plans it's an add-on with no public price; on the PD Reliability Platform it's included from the Essential tier at $12,000 a year. Its SRE agent is an add-on on Starter and included from Essential. It's best for teams already paging through PagerDuty, and it doesn't replace your monitoring.

Pricing
AIOps add-on: no public price; Incident Management Professional $21 per user per month billed annually ($25 monthly); PD Reliability Platform from $2,800 per year (Starter), with Signal Intelligence included from Essential ($12,000 per year) (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS
Runs on
Hosted; takes events from your monitoring tools

Pros

  • Mature noise reduction and event orchestration
  • Works across many monitoring sources
  • SRE agent available on higher tiers

Cons

  • AIOps add-on price isn't published
  • Platform tiers start at $2,800 a year
  • Needs PagerDuty as your incident hub

Not tested by us yet: based on the vendor's documentation.

incident.io AI SRE

Best for: AI investigation inside your incident process

incident.io's AI SRE starts investigating when an incident is declared, linking telemetry, code changes and past incidents to a root cause with a confidence score and the sources behind each finding. Its safety model is clear: the only change it can make to your systems is a pull request you review and merge. Each customer gets an isolated data instance, and sensitive data is redacted before it reaches model providers. The AI agent is included on Pro, at $25 per user per month, and on Enterprise; root cause analysis and the autonomous agent are add-ons without public prices. It's best if you'd also run incidents in incident.io.

Pricing
Basic free; AI agent included on Pro ($25 per user per month) and Enterprise; root cause analysis and the autonomous agent are add-ons with no public price (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS
Runs on
Hosted; works inside incident.io incident response

Pros

  • Changes systems only through pull requests you merge
  • Shows confidence and sources for findings
  • Part of a full incident-response product

Cons

  • Key AI features are add-ons with no public price
  • Not available on the Basic or Team plans
  • Best value only if you adopt incident.io

Not tested by us yet: based on the vendor's documentation.

Cleric

Best for: a standalone AI SRE with published pricing

Cleric checks each deploy for regressions for 14 days, triages alerts by closing false positives and escalating real ones with evidence, and investigates issues through to a proposed fix pull request. It's read-only by default, logs every action, and works in Slack and Microsoft Teams, or over MCP from coding agents. Unusually, its prices are public: Team is $700 a month, and work is billed in $0.50 credits, so $1 per change verified and $5 per issue investigated, with 1,000 trial credits. That's a lot for one or two servers, but easy to budget for a product team that ships daily.

Pricing
Team $700, Pro $1,500 and Scale $3,000 per month; credits cost $0.50 ($1 per change verification, $5 per issue investigation); 1,000 trial credits (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS, with a connector for private resources
Runs on
Hosted; connects to your repos, monitoring and chat; read-only by default

Pros

  • Read-only by default, with every action logged
  • Published pricing per unit of work
  • Checks deploys for regressions, not just alerts

Cons

  • Starts at $700 a month
  • Aimed at teams shipping code often

Not tested by us yet: based on the vendor's documentation.

Resolve AI

Best for: large teams that want an agent on primary on-call

Resolve AI's pitch is machines taking on-call for humans. Its agents take primary on-call, triage and mitigate alerts, investigate root causes, and escalate to engineers when needed, and it can build custom SRE agents for other operational work. It lists SAML SSO, RBAC, encryption, redaction, retention controls and customer data isolation, and says customer data isn't used to train models for others. The site cites customers such as Coinbase and Zscaler. There's no public price, and the homepage doesn't describe the deployment model. Shortlist it for large on-call rotations; it isn't aimed at a few servers.

Pricing
No public price (contact sales) (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS
Runs on
Hosted; works with your existing observability, code and incident tools

Pros

  • Aims to take on-call work end to end
  • Security controls listed, including redaction and data isolation
  • Custom agents for other operational tasks

Cons

  • No public price
  • Deployment model not described on its homepage
  • Built for large organizations

Not tested by us yet: based on the vendor's documentation.

Rootly AI SRE

Best for: teams that already run incidents in Rootly

Rootly AI SRE adds agents to Rootly's incident platform that find a probable root cause with evidence, correlate alerts with recent deploys and configuration changes, suppress noise, and draft remediation steps and pull requests. It includes MCP and AI connectors and works with tools such as Claude and Codex. AI SRE is priced through sales and requires Rootly Incident Response, which costs $20 per user per month; Rootly offers bundle discounts and up to 50% off for qualifying startups. Pick it if Rootly is, or will be, your incident hub.

Pricing
AI SRE: contact sales (requires Incident Response at $20 per user per month) (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS
Runs on
Hosted; requires Rootly Incident Response

Pros

  • Correlates alerts with deploys and config changes
  • Drafts remediation steps and pull requests
  • MCP connectors

Cons

  • AI SRE price isn't published
  • Requires Rootly Incident Response

Not tested by us yet: based on the vendor's documentation.

Dynatrace

Best for: enterprises that want AI inside full-stack observability

Dynatrace builds AI into its observability platform instead of selling it as a separate tool. AI exploration, covering anomaly detection and log pattern analysis, is included at no extra charge, and Agentic Operations, which includes autonomous investigations, is billed at $0.10 per AI unit. Full-Stack Monitoring costs $58 a month per 8 GiB host and Infrastructure Monitoring $29 per host, with logs at $0.20 per GiB ingested. It's a strong choice when you want one vendor for monitoring and AI analysis, and you get the most from it when Dynatrace is also your main monitoring.

Pricing
Full-Stack Monitoring $58 per month per 8 GiB host; Infrastructure Monitoring $29 per host per month; logs $0.20 per GiB ingested; AI exploration included; Agentic Operations $0.10 per AI unit (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS with an agent on each host
Runs on
Hosted; monitors hosts, applications, logs and cloud services

Pros

  • AI exploration included in the platform
  • Published usage prices
  • Monitoring and AI analysis from one vendor

Cons

  • Works best when Dynatrace is your main monitoring
  • Usage-based pricing needs careful forecasting

Not tested by us yet: based on the vendor's documentation.

BigPanda

Best for: large IT operations teams drowning in alerts

BigPanda is an agentic AI platform for IT operations. It turns raw alerts into context-rich incidents, cuts ticket volume, and offers AI help for L1 and L2 operators, through products such as AI Detection & Response, AI Incident Assistant and an IT Knowledge Graph. It's built for large enterprises with many monitoring tools and a service desk to feed. There's no public price; its site offers a return-on-investment claim rather than plans. For a small team running Linux servers it's more platform than you need; PagerDuty's AIOps or an open-source agent is a lighter start.

Pricing
No public price (contact sales) (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS
Runs on
Hosted; takes alerts from monitoring, observability and service-desk tools

Pros

  • Strong focus on alert correlation and noise
  • Service-desk correlation
  • Assistance for L1 and L2 operators

Cons

  • No public price
  • Enterprise-scale product

Not tested by us yet: based on the vendor's documentation.

Komodor

Best for: Kubernetes-heavy teams

Komodor is an agentic operations platform centered on Kubernetes. Its incident module groups related alerts into one incident, sends specialist agents to gather evidence in parallel, writes a root-cause conclusion with that evidence attached, and stages a fix that a human must approve before anything reaches production. A remediation posture sets what may be staged: nothing beyond a conclusion, pre-approved runbook steps, or an action a named remediation agent composes, with approval required in every case. Agents can run on Komodor's infrastructure or yours, and a self-hosted control plane is available under agreement (in beta), not as a self-service install. There's no public price: its pricing page lists custom pricing made of a platform fee plus AI tokens, with a trial. If your servers aren't on Kubernetes, look elsewhere.

Pricing
No public price: custom pricing made of a platform fee plus AI tokens; free trial (checked November 15, 2026, source)
License
Proprietary
Deployment
SaaS, with optional self-hosted agents; a self-hosted control plane by agreement (beta)
Runs on
Kubernetes-focused; its agents also read AWS, Azure, Google Cloud, Datadog and Grafana

Pros

  • Human approval gate before any fix
  • Evidence attached to every conclusion
  • Agents can run in your own cluster; self-hosted control plane by agreement

Cons

  • No public price
  • Built around Kubernetes

Not tested by us yet: based on the vendor's documentation.

K8sGPT

Best for: a free first look at what's wrong in a Kubernetes cluster

K8sGPT scans a Kubernetes cluster with built-in analyzers and explains the problems it finds in plain English. It works with OpenAI, Azure, Amazon Bedrock, Google Gemini, Cohere and local models such as Ollama, and an --anonymize flag masks names before data goes to the AI backend, though its README lists analyzers that aren't masked (Pod, ReplicaSet, PVC, Log and Events), and event messages can still carry names. It can run as an in-cluster operator and as an MCP server for AI assistants. It's Apache-2.0 and free, apart from any hosted model you use. It's a diagnostic helper, not an incident system, and it does nothing for plain Linux servers.

Pricing
Free to use under its open-source license (Apache-2.0); you pay for any hosted AI backend (checked November 15, 2026, source)
License
Open source (Apache-2.0)
Deployment
Self-hosted CLI or in-cluster operator
Runs on
Kubernetes clusters; AI backends include OpenAI, Azure, Amazon Bedrock, Google Gemini, Cohere and local models

Pros

  • Free and open source
  • Works with local models for privacy
  • MCP server mode for AI assistants

Cons

  • Kubernetes only
  • Explains problems; doesn't investigate incidents end to end

Not tested by us yet: based on the vendor's documentation.

How we picked

We read each product's own pages, documentation and pricing on November 15, 2026, and confirmed that each product exists and is offered today; we didn't run these tools ourselves. We ranked them on:

  • Real investigation: does it find a likely cause and show the evidence, or only group alerts?
  • Safety: read-only by default, human approval before changes, and a record of what it did.
  • Fit with your data: whether it works with the monitoring and logs you already have.
  • Price and access: published prices and a way for a small team to start.
  • Data handling: redaction, isolation, and whether your data trains anyone's models.

We didn't run these tools: judging them fairly needs a production observability stack, an incident platform or a Kubernetes cluster, which our small test servers don't have. Every claim comes from the vendor pages listed under Sources, and vendor performance claims are labeled as theirs. Nobody paid to be listed. We make HostCrust, a server operations product, and it isn't ranked here.

AIOps or AI SRE: what's the difference?

AIOps is the older idea: apply machine learning to the flood of alerts and events. It groups related alerts into one incident, suppresses duplicates, detects anomalies and hints at a probable origin. PagerDuty AIOps, BigPanda and Dynatrace's built-in AI work mainly at this level, and they're most valuable when many monitoring tools page the same people.

An AI SRE is newer: an agent built on a large language model that investigates like an engineer. It queries metrics and logs, reads recent deploys and code, forms hypotheses and returns a root cause with evidence, sometimes with a proposed fix. HolmesGPT, Datadog Bits Investigation, Cleric, Resolve AI, incident.io, Rootly and Komodor work this way. K8sGPT sits in between: it explains what its analyzers find in a cluster rather than running a full investigation.

What should you look for in an AI SRE tool?

Start with what the agent may touch. The safest designs are read-only by default and need a human to approve any change, or limit changes to pull requests you review. Next, demand evidence: a root cause is only useful if you can see the queries, log lines and deploys behind it, plus a confidence level, because a confident wrong answer costs more time than no answer. Check where your data goes, since agents read logs that may hold customer data and secrets: look for redaction, per-customer isolation, and a promise that your data won't train shared models. Make sure it reads the tools you already use, such as Prometheus, Loki, Datadog or your incident platform, rather than requiring a migration. Finally, understand the cost model. Per-investigation credits, per-seat add-ons and flat monthly plans behave very differently when a bad night fires hundreds of alerts.

Safety and pricing at a glance

ToolCan it change your systems?How it's priced
HolmesGPTRead-only by design; optional toolsets can scale, roll back or open PRsFree; you pay your model provider
Datadog Bits InvestigationNot described on its product page; it investigates and reportsAI credits, about 4 per investigation
PagerDuty AIOpsEvent orchestration automates actions you defineAdd-on, or included from Essential ($12,000 a year)
incident.io AI SREOnly through pull requests you mergeIncluded on Pro; some features are add-ons
ClericRead-only by default; write access optionalFrom $700 a month; $0.50 credits
Resolve AIMitigates alerts; method not described on its homepageQuote
Rootly AI SREDrafts remediation steps and PRsQuote; needs Incident Response ($20 per user)
DynatraceNot described on its pricing page$0.10 per AI unit for Agentic Operations
BigPandaNot described on its homepageQuote
KomodorStages fixes behind a human approvalQuote
K8sGPTReads and explainsFree; you pay any hosted model

Open source or SaaS?

The open-source agents, HolmesGPT and K8sGPT, run where you choose, and use the model provider you choose (K8sGPT also supports local models), so you decide what data reaches which provider. You do the setup, choose the credentials and tune the prompts and toolsets. The SaaS tools add things that are hard to build yourself: memory of past incidents, ready integrations, deploy and code context, and someone to call when it misbehaves. They also see your telemetry, so read their data-handling terms as carefully as their feature lists.

Do you need AIOps or an AI SRE at all?

For a handful of Linux servers, probably not yet. These tools reason over your telemetry, so without good server monitoring, centralized logs and sensible alerts, they have little to work with and will guess. Get those right first. Once you have them, a free agent such as HolmesGPT pointed at your existing Prometheus and Loki is a low-risk way to see whether AI investigation saves you time. The paid platforms make sense when alert volume, team size or on-call cost is large enough to measure the savings.

How we checked this: Researched from vendor documentation, product pages and pricing pages checked on 15 Nov 2026; none of these tools were run on our servers. They need a production observability stack, an incident platform or a Kubernetes cluster to judge fairly, which our two small test servers don't have. Anything we didn't run ourselves comes from the official documentation listed under Sources.

Give AI agents SSH with a seatbelt

ssh-mcp-guard lets an AI agent read freely but makes every change wait for a signed human approval, enforced on the server. Free and open source.

About ssh-mcp-guard

Check which CVEs are really open on your server

HostCrust Patch Proof compares every installed package with your distribution's own security data. It counts fixes your distro backported and checks the kernel that is actually running, not just the one on disk. Read-only, no root, free.

Run the free check About Patch Proof

FAQ

What is AIOps?

AIOps applies machine learning to IT operations data, mostly alerts and events: grouping related alerts, suppressing noise, spotting anomalies and pointing to likely causes. PagerDuty AIOps, BigPanda and Dynatrace's built-in AI are typical examples.

What is an AI SRE?

An AI SRE is an agent, usually built on a large language model, that investigates an alert the way an engineer would: it reads metrics, logs, recent deploys and code, then reports a likely root cause with evidence and sometimes a proposed fix. HolmesGPT, Datadog Bits Investigation, Cleric, Resolve AI, incident.io and Rootly all offer one.

Are there open-source AIOps or AI SRE tools?

Yes. HolmesGPT (Apache-2.0, a CNCF sandbox project) investigates alerts across Prometheus, Loki, Grafana, Datadog, Kubernetes and more. K8sGPT (Apache-2.0) scans Kubernetes clusters and explains problems. Both let you choose the model provider, and K8sGPT supports local models.

Is it safe to let an AI agent change production?

Only with guardrails. Prefer tools that are read-only by default (HolmesGPT, Cleric), change systems only through pull requests you merge (incident.io), or stage fixes behind a human approval (Komodor). Give agents the narrowest credentials that work, and keep a log of everything they did.

How much do AI SRE tools cost?

HolmesGPT and K8sGPT are free apart from model costs. Datadog Bits Investigation works out to roughly $4 per investigation at its annual credit rate, by our math. Cleric starts at $700 a month. Most others, including Resolve AI, BigPanda, Komodor, Rootly AI SRE and PagerDuty's AIOps add-on, are priced by quote.

Sources

  1. HolmesGPT on GitHub (CNCF sandbox, read-only design, toolsets, Apache-2.0)
  2. K8sGPT on GitHub (analyzers, backends, anonymize, MCP)
  3. Datadog Bits Investigation
  4. Datadog pricing (AI credits)
  5. PagerDuty pricing (AIOps, Signal Intelligence, SRE agent)
  6. PagerDuty AIOps
  7. incident.io AI SRE
  8. incident.io pricing
  9. Cleric
  10. Cleric pricing
  11. Resolve AI
  12. Rootly pricing (AI SRE)
  13. Dynatrace pricing
  14. BigPanda
  15. Komodor docs: incident management and troubleshooting
  16. Komodor docs: deployment methods
  17. Komodor pricing and plans (custom pricing: platform fee plus AI tokens)