Datadog Bits AI alternatives: 10 AI SRE agents compared on data, actions and pricing

Bits AI SRE is now Bits Investigation, and Datadog has replaced per-investigation pricing with AI Credits shared across its Bits agents. Datadog estimates 6.5 credits for an autonomous investigation. That isn't a fixed charge, and the alternatives don't necessarily measure the same work. Comparing the price of a credit won't tell you which agent costs less to investigate an incident.

Some agents can query Datadog directly. Others need the telemetry in their own platform, which means moving it or sending it to two places. And investigating an incident is not the same as coordinating the response, proposing a fix, or being allowed to execute one in production.

What Bits Investigation does today

Bits Investigation uses Datadog data and context to investigate alerts, returning either an evidence-backed conclusion or an inconclusive result.

Triggers. Auto-Investigate is configured individually on each monitor and fires when that monitor enters Alert. Warn, no-data, renotify, and test notifications don't start an investigation. Neither the API nor Terraform can enable Auto-Investigate. For a feature meant to automate investigations, rolling it out across hundreds of monitors involves rather a lot of manual work. Investigations can also be launched from a monitor, Synthetics, RUM, a free-form prompt, or @Datadog in Slack. Supported monitor types include metric, anomaly, forecast, logs, APM (metrics only), composite, SLO, Synthetics, and RUM (Datadog investigation docs).

Data. Bits uses Datadog metrics, traces, logs, dashboards, events, Change Tracking, Watchdog, RUM, Network Path, Database Monitoring, and Profiler data. GitHub provides source-code context. The Grafana, Dynatrace, Sentry, Splunk, ServiceNow, and Confluence integrations are in Preview. Datadog accepts telemetry through OTLP, but sending OpenTelemetry data into Datadog is not the same as having Bits investigate data held exclusively in another backend. The reviewed documentation does not establish that latter capability.

Memory and actions. Feedback on conclusions can be retained as user-viewable and deletable memory. The bits.md guidance file, instructions in monitor messages, and Confluence runbooks can influence investigations (knowledge sources). Triage actions—Slack or Teams messages, pages, incidents or cases, and Jira tickets—require human approval. Rollbacks and other remediation workflows can use Datadog's Action Catalog. Bits Code, Chat, and Security Analyst are separate agents. Replacing Investigation won't replace the rest of the Bits family.

AI Credit products aren't available to FedRAMP and US-FED customers (Datadog billing docs). Bits memory and guidance remain in Datadog even if you migrate the underlying telemetry.

Bits now draws from a shared credit pool

AI Credits cost $1.30 each on demand. The committed plan starts at $500 for 500 monthly credits, billed annually. Unused monthly credits do not roll over, so a bad usage forecast can leave you paying for credits you never use. Datadog estimates 6.5 credits for an autonomous Bits investigation. A Bits Code fix uses roughly five credits and a Chat message about 0.5, according to Datadog's averages. They share a pool, while telemetry ingestion and indexing are billed separately (Datadog AI Credit pricing).

An investigation's actual credit use varies, and Code and Chat spend from the same allocation. You can set organization and user caps—useful when Bits is enabled by default for the Standard role. Datadog also offers a 14-day trial.

Datadog's published AI Credit terms do not specify whether inconclusive investigations consume credits.

Where AI SRE agents get their data

Platform-native agents investigate telemetry available through their own observability platforms. Bits works on Datadog data; Agent0, Grafana Assistant Investigations, and New Relic Autopilot use data available in their respective platforms. If Datadog remains your backend, testing Agent0 or Grafana on the same telemetry means sending it to a second platform or migrating it.

Standalone agents such as Resolve AI, Cleric, and Traversal query Datadog and other backends through integrations and API credentials. The telemetry can stay where it is. The agent still needs credentials and has to work within the backend's API limits. Check what the provider does with query results, incident context, and generated artifacts. Read-only access limits what an agent can change, not what it can see.

Agents attached to incident, on-call, or cloud platforms, including incident.io, PagerDuty, Rootly, and AWS, connect investigations to existing incident or cloud workflows. Their scope is limited by the data those integrations expose.

A standalone agent avoids a telemetry migration, not the work of granting another vendor access. Choosing another native platform brings a different problem: the telemetry must reach it, while the agent's memories and workflows stay tied to that platform.

Ten alternatives at a glance

Platform-native agents

Dash0 Agent0

Dash0 Agent0 investigates telemetry in Dash0. It can also generate dashboards and alert rules and prepare pull requests for code, instrumentation, or configuration fixes. Automations trigger configured workflows on events or schedules, with permissions and guardrails controlling what the agent may do (Agent0).

Agent0 credits cost $0.60 each. A complex task, including root-cause analysis or preparing a pull request, typically uses 1–2 credits; telemetry is billed separately. Agent0 uses foundation models hosted through AWS Bedrock and Google Cloud Vertex AI under zero-data-retention agreements. Dash0 offers EU telemetry storage residency and configurable model capability levels.

Agent0 can't investigate telemetry retained only in Datadog. To use it, you need to send the relevant telemetry to Dash0. Agent0 doesn't provide on-call or incident management, so those workflows still need another tool. Datadog also has a broader integration catalog than Dash0.

Grafana Assistant Investigations

Grafana Assistant Investigations runs hypotheses against metrics, logs, traces, profiles, deployments, and alerts. Investigations can start from an alert enricher, IRM webhook, Slack, MCP, or a manual request (Grafana). The agent recommends steps; it does not execute them.

Grafana doesn't charge per investigation. The free tier includes three AI users and a 25-million-token system pool for automatic investigations. Grafana Cloud Pro costs $19/month as a platform fee, plus $20 per active AI user with 40 million tokens, and $2 per additional million tokens (Grafana pricing). Investigations require Grafana Cloud; the feature is unavailable in self-managed Grafana.

New Relic Autopilot

Autopilot uses traces, logs, metrics, deployments, specialist coverage for Kubernetes and Kafka, team memory, and GitHub/Jira context over MCP. It recommends remediations rather than executing them. It requires Pro or Enterprise, limits usage to 100 requests per hour per organization, and isn't available for FedRAMP (New Relic docs). Public AI pricing isn't provided in the reviewed sources.

New Relic's blog and documentation disagree about whether Advanced Compute is required and whether alert-triggered investigations are generally available. The current overview describes on-demand chat. That's a poor uncertainty to discover halfway through an automated rollout: confirm the entitlement and trigger support with New Relic first.

Standalone agents over existing backends

Resolve AI

Resolve advertises integrations with more than 60 tools, including observability, incident, and source systems. Connecting Datadog means providing an API key and an App key with read scopes, then sending alerts through @webhook-resolveai (Datadog integration). Mitigation requires additional write permissions. That changes the agent's ability to affect production, not merely what appears in its feature list. Self-managed Grafana needs a Helm-deployed Satellite. Resolve pricing is not public.

Cleric

Cleric takes the pager, investigates alerts, and verifies deployments for up to 14 days after a change. It is read-only by default and integrates with Datadog, Grafana, Prometheus, New Relic, Splunk, Elastic, Honeycomb, Sentry, PagerDuty, incident.io, GitHub, and generic MCP. Cleric reports SOC 2 Type II (integrations).

Cleric publishes its pricing. The Team plan is $700/month for 1,400 credits, and an investigation consumes 10 credits: $5 at $0.50/credit, or $4 at its stated $0.40 annual-plan rate. Cleric says it won't bill overages without written approval (pricing).

Traversal

Traversal uses causal search over what it calls a “Production World Model.” Its agentless deployment is read-only by default, and it offers bring-your-own-cloud (BYOC) and bring-your-own-model options (Traversal). Traversal lists Datadog, Splunk, New Relic, Elastic, and Prometheus as sources. The reviewed research didn't include a detailed integration catalog, so confirm which endpoints and permissions are supported before relying on the integration. Pricing is not public.

Agents attached to incident, on-call or cloud platforms

incident.io

Declaring an incident in incident.io can start an investigation that returns a confidence score and citations. An adversarial agent challenges the initial conclusion (incident.io). Datadog and Grafana connections are read-only. incident.io describes dedicated instances per customer, zero-data-retention agreements with LLM providers, and SOC 2 compliance. Investigations are a Pro/Enterprise add-on with no published AI price.

PagerDuty SRE Agent

PagerDuty SRE Agent is available in Incident Details; escalation-policy triggers remain in early access in the reviewed material. It reads Datadog logs through the API and metrics over MCP and can propose GitHub pull-request fixes. Other connections include Grafana, New Relic and Splunk. The agent is included in Reliability Platform plans. Starter costs $2,800/year and includes 1,000 AI Actions (pricing). PagerDuty doesn't publish how many actions an investigation or proposed fix consumes, so the allowance doesn't tell you how many investigations you're buying.

Rootly

Rootly starts an AI SRE investigation approximately 30 seconds after an alert, runs hypotheses in parallel, and remains read-only. Integrations include Datadog, Grafana, Sentry, and GitHub (Rootly). The AI agent is a quoted add-on to Rootly Incident Response, which starts at $20/user/month. The AI-investigation price is not public.

AWS DevOps Agent

AWS DevOps Agent connects to CloudWatch, Datadog, Dynatrace, Grafana, New Relic, and Splunk. It charges by elapsed agent time: $0.0083 per agent-second, or roughly $30 per agent-hour, plus charges for the AWS services it queries (pricing). That's a rate, not an investigation price. Without measuring how long the agent runs on your incidents, you don't know what those investigations will cost.

Why the published prices don't rank the agents

Datadog's reference estimate is 6.5 credits per autonomous Bits investigation at $1.30 per credit on demand. An Agent0 complex task typically uses 1–2 credits at $0.60 per credit. Applying those estimates to a hundred runs gives:

Those numbers describe different vendor-defined tasks, and neither is a fixed fee. The examples also exclude telemetry, contracts, unused committed credits, and migration costs.

The arithmetic works. The comparison doesn't.

Run both agents on the same resolved incidents, measure what each actually consumes, and compare credit caps and overage policies.

Shortlisting by the constraint you actually have

When telemetry has to stay in Datadog, begin with Bits as the native baseline. Cleric publishes pricing and defaults to read-only; Traversal offers BYOC and model control; Resolve offers broad integrations and mitigation actions with additional permissions. If you already use incident.io, PagerDuty, or Rootly, evaluate its agent before introducing another service.

For predictable AI spend, look at what the billing unit covers and whether the vendor lets you cap it. Datadog offers user and organization credit caps. Cleric publishes a ten-credit investigation and says it won't bill unapproved overages. PagerDuty uses AI Actions and Grafana uses tokens; neither is directly comparable to a per-investigation credit estimate. An average is not a spending cap.

For OpenTelemetry portability, Agent0 uses telemetry in Dash0 but doesn't replace on-call. Datadog can ingest OTLP while you continue using Bits Investigation. OpenTelemetry can make instrumentation and collection more portable; it doesn't move Bits memory, bits.md, or platform-specific workflows to a replacement agent. For related detail, see Datadog alternatives and migrating from Datadog to OpenTelemetry and Dash0. For a general ranking outside the Datadog context, see best AI SRE tools.

Test with incidents you've already resolved

Use closed incidents with a known root cause, and run those same incidents through the shortlisted agents. Published vendor benchmarks aren't a substitute for that test. Track the answers, inconclusive outcomes, time taken, human work still required, and the actual credits, tokens, AI Actions, or agent-seconds consumed. incident.io recommends more than 80% precision—correct conclusions divided by all conclusions made—before relying on an agent. The threshold is incident.io's recommendation, not an industry standard.

Then check what the agent can do to production. Start read-only; record API scopes and rate limits. Add write access only for remediations you've approved. An investigation should leave an audit trail of what was queried, concluded, and executed. Verify LLM subprocessors, retention, training terms, residency, and BYOC/BYOM options where relevant. Check spend caps or overage approvals, and find out whether memory, runbooks, guidance files, and generated artifacts can be exported when you leave.

Datadog reports 90% faster service restoration, while Cleric reports 58% autonomous alert closure. Restoration time and autonomous closure rate are not the same metric. Treating them as a common accuracy ranking would be nonsense. Compare agents on the same incidents instead.

Final thoughts

Replacing Bits Investigation is one decision; replacing Datadog as the telemetry backend is another. Resolve, Cleric, and Traversal let you investigate data kept in Datadog, with the API permissions and data-handling obligations that come with it. Agent0 needs telemetry in Dash0; Grafana Assistant needs it in Grafana Cloud. An agent attached to PagerDuty or incident.io changes the incident workflow rather than the underlying observability stack.

Don't choose from accuracy percentages measuring different outcomes or from credit-price arithmetic comparing different tasks. Run incidents you've already resolved, compare the conclusions, and measure how much investigation work remains for an engineer—as well as what the agents actually cost. If you're considering an OpenTelemetry-native backend, try Agent0 with representative telemetry.