The best AIOps and AI SRE tools in 2026
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
- HolmesGPTBest for a free, open-source AI investigator you control
- Datadog Bits InvestigationBest for teams whose telemetry already lives in Datadog
- PagerDuty AIOpsBest for cutting alert noise across many monitoring tools
Compare at a glance
| Tool | Best for | Pricing | Open source | Deployment | Tested by us |
|---|---|---|---|---|---|
| HolmesGPT | a free, open-source AI investigator you control | Free (you pay your LLM provider) | Yes | Self-hosted CLI, server or operator | Docs only |
| Datadog Bits Investigation | teams whose telemetry already lives in Datadog | 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 | No | SaaS (part of Datadog) | Docs only |
| PagerDuty AIOps | cutting alert noise across many monitoring tools | 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) | No | SaaS | Docs only |
| incident.io AI SRE | AI investigation inside your incident process | 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 | No | SaaS | Docs only |
| Cleric | a standalone AI SRE with published 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 | No | SaaS, with a connector for private resources | Docs only |
| Resolve AI | large teams that want an agent on primary on-call | No public price (contact sales) | No | SaaS | Docs only |
| Rootly AI SRE | teams that already run incidents in Rootly | AI SRE: contact sales (requires Incident Response at $20 per user per month) | No | SaaS | Docs only |
| Dynatrace | enterprises that want AI inside full-stack observability | 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 | No | SaaS with an agent on each host | Docs only |
| BigPanda | large IT operations teams drowning in alerts | No public price (contact sales) | No | SaaS | Docs only |
| Komodor | Kubernetes-heavy teams | No public price: custom pricing made of a platform fee plus AI tokens; free trial | No | SaaS, with optional self-hosted agents; a self-hosted control plane by agreement (beta) | Docs only |
| K8sGPT | a free first look at what's wrong in a Kubernetes cluster | Free to use under its open-source license (Apache-2.0); you pay for any hosted AI backend | Yes | Self-hosted CLI or in-cluster operator | Docs 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
| Tool | Can it change your systems? | How it's priced |
|---|---|---|
| HolmesGPT | Read-only by design; optional toolsets can scale, roll back or open PRs | Free; you pay your model provider |
| Datadog Bits Investigation | Not described on its product page; it investigates and reports | AI credits, about 4 per investigation |
| PagerDuty AIOps | Event orchestration automates actions you define | Add-on, or included from Essential ($12,000 a year) |
| incident.io AI SRE | Only through pull requests you merge | Included on Pro; some features are add-ons |
| Cleric | Read-only by default; write access optional | From $700 a month; $0.50 credits |
| Resolve AI | Mitigates alerts; method not described on its homepage | Quote |
| Rootly AI SRE | Drafts remediation steps and PRs | Quote; needs Incident Response ($20 per user) |
| Dynatrace | Not described on its pricing page | $0.10 per AI unit for Agentic Operations |
| BigPanda | Not described on its homepage | Quote |
| Komodor | Stages fixes behind a human approval | Quote |
| K8sGPT | Reads and explains | Free; 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.
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.
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
- HolmesGPT on GitHub (CNCF sandbox, read-only design, toolsets, Apache-2.0)
- K8sGPT on GitHub (analyzers, backends, anonymize, MCP)
- Datadog Bits Investigation
- Datadog pricing (AI credits)
- PagerDuty pricing (AIOps, Signal Intelligence, SRE agent)
- PagerDuty AIOps
- incident.io AI SRE
- incident.io pricing
- Cleric
- Cleric pricing
- Resolve AI
- Rootly pricing (AI SRE)
- Dynatrace pricing
- BigPanda
- Komodor docs: incident management and troubleshooting
- Komodor docs: deployment methods
- Komodor pricing and plans (custom pricing: platform fee plus AI tokens)