Bits AI
A family of generative AI features that lets you question the monitoring data stored in Datadog in natural language and delegate alert investigation, code fixes, and first-pass security triage. This article draws the line between what to delegate and where a human decides, within the scope of the official documentation.
What it does
In Bits Chat you ask in natural language, such as "what caused the 400 errors in the last hour," and it searches across metrics, logs, and traces. You hand alert investigation to Bits Investigation to gather hypotheses and evidence, and have Bits Code produce proposed code fixes as pull requests. Cloud SIEM signals are investigated by Bits Security Analyst, which proposes a verdict. Billing is measured in a unit called AI Credits.
When to use it
It suits the on-call engineer who wants to spend the first 15 minutes of a late-night alert enumerating hypotheses, the developer who wants to trace the change that caused an incident down to the code and get a draft pull request, the security staff who want a first-pass verdict on a large volume of security signals before a human looks, and the user who wants to ask "show me" before learning how to build a dashboard.
An everyday analogy
It is a hotel concierge. Ask "which nearby restaurants are open after 10 pm" and the concierge checks the in-house materials and outside reservation systems, lines up candidates with reasons, and will even make the reservation if asked. But the guest decides which restaurant to go to, and the concierge never exceeds the range of reservations that can be made in the guest's name.
Conclusion - telling apart three AIs that point in different directions
What Datadog calls "AI" splits into three directions. Bits AI is the family of generative AI features that lets users advance an investigation through dialogue or delegate whole tasks such as investigation and fixes. The official documentation positions it as "an agentic teammate inside Datadog, built to automate development, security, and operations workflows," and explains that you can converse with it in real time or entrust it with work such as alert investigation, code fixes, and security triage (as of September 2026; source: Bits AI). LLM Observability points the opposite way: it is the side that monitors the LLM applications you build yourself. Watchdog is built-in anomaly detection that works without configuration and runs even when nobody asks it anything. Pricing is measured in a unit called AI Credits, but because the unit price is not on any SKU row of the public pricing table, this article does not give numbers.
| Name | Direction | How people are involved |
|---|---|---|
| Bits AI | Advance an investigation through dialogue against Datadog's monitoring data, or delegate investigation and fixes | Ask, delegate, verify the results |
| LLM Observability | Monitor the traces, tokens, and evaluations of your own LLM applications | Instrument and observe (the monitored party is your own AI) |
| Watchdog | Automatically detect anomalies in metrics and logs without configuration | Receive results (no asking) |
Seven components, three of them in Preview
As of September 2026, the Features section of the official documentation lists seven features (source: Bits AI). Bits Investigation is an autonomous agent that investigates production issues from start to finish; it forms hypotheses, gathers telemetry, and helps the on-call engineer identify the root cause through data-driven reasoning (source: Bits Investigation). Bits Code is a generative AI coding assistant that diagnoses and fixes code problems from monitoring data; it integrates with source code management services to create pull requests or merge requests and iterates on fixes in response to CI logs and developer feedback (source: Bits Code). Bits Security Analyst investigates Cloud SIEM signals and issues verdict recommendations (source: Bits Security Analyst). Bits Chat is the natural-language entry point, and the remaining three are marked Preview on their official pages.
| Feature | Official description (summarized) | Availability |
|---|---|---|
| Bits Chat | Search across and operate Datadog in natural language | Available in the web app, mobile app, and Slack |
| Bits Investigation | Autonomously investigate production issues and help identify the root cause | Available |
| Bits Code | Diagnose code problems from monitoring data and create pull requests | Available |
| Bits Security Analyst | Investigate Cloud SIEM signals and recommend a verdict | Available |
| Bits Data Analysis | Explore business data in natural language | Preview |
| Bits Detection | Autonomously monitor for high-impact service degradations | Preview |
| Bits Remediation | Take action against the root cause | Some features in Preview |
The entry point is Bits Chat on web, mobile, and Slack
The everyday entry point is Bits Chat. The official documentation describes it as a feature to "search across and operate Datadog in natural language" and states that it can be used in the web app, the mobile app, and Slack (as of September 2026; source: Bits Chat). The example questions on that page are investigative ones such as "summarize the high-severity incidents of the past day" and "why is the checkout endpoint returning 400 errors in the last hour," and Bits pulls in monitors, logs, and traces as evidence. Because the scope of the question determines the scope of the answer, a question that includes the service name, the time window, and the symptom pins down what to investigate, while a question that says only "something is off" does not. The significance of using it from Slack is that you can throw the first question without switching screens, right where the incident-response conversation is happening, so the on-call engineer's act of writing up the situation becomes the starting point of the investigation.
The billing unit is AI Credits; see the public pricing table for the numbers
Bits AI billing is measured in AI Credits. The official documentation defines "1 AI Credit as one unit of intelligent work performed by Datadog's AI products" and lists Bits Chat, Bits Investigation, Bits Code, and Bits Agent Builder as the products that consume them (as of September 2026; source: AI Credits). The pricing tiers and administrative controls are shown in the figure below; the unit price is not on any SKU row of the public pricing table, so this article does not give it. Unlike per-host or per-GB prices, consumption of a "unit of intelligent work" varies widely with how it is used, so the safe order is to set per-user limits first and decide the organization-wide limit after observing actual consumption. For how to read the pricing structure as a whole, see how to read the pricing structure. This article explains how billing works. For current unit prices and plans, refer to Datadog's official pricing page.
- Annual Commit: a 12-month commitment, billed monthly, credits purchased in bundles
- Monthly Commit: a month-by-month contract, credits purchased in bundles
- On-Demand: no commitment, billed monthly based on actual usage
- Administration: monthly credit limits per organization and per user, disabling of specific products
External AI agents connect through the MCP Server
Its counterpart is the Datadog MCP Server. If Bits AI is the AI inside Datadog, the MCP Server is the side that bridges external AI agents (coding assistants and chat clients) to Datadog's data; the details are left to the glossary entry for MCP Server. Permissions and data handling are stated explicitly in the official documentation. Because the MCP Server passes the authenticated user's own credentials to the Datadog API, existing access controls apply as-is, and it cannot reach resources the user cannot see in the UI. It is the user's AI client that sends Datadog data to a third-party AI provider, and that flow follows the user's own agreements. Most tool calls query Datadog directly without going through a model; only some, such as semantic search, use a model from Datadog's AI provider, and these can be disabled (as of September 2026; source: Datadog MCP Server).
| Aspect | What the official documentation states |
|---|---|
| Permissions | Forwards the user's own credentials. Existing controls such as RBAC apply as-is, and access beyond the user's permissions is not possible |
| What it receives | Only tool names and arguments. It does not receive prompts or model reasoning |
| Outbound transmission | The MCP Server does not send to external domains. What is sent to the AI provider is decided by the user's client |
| Use of AI models | Most tools do not go through a model. Only some, such as semantic search, use a model from Datadog's AI provider, and these can be disabled |
| Records | Tool calls can be traced in the Audit Trail |
| Limits and eligibility | Bursts of 50 requests per 10 seconds. HIPAA-eligible; GovCloud is not supported |
What to delegate, and what a human decides
In the official wording, what Bits AI produces is "hypotheses and evidence," "a pull request with a proposed fix," and "a recommended verdict." Bits Investigation "helps" identify the root cause, Bits Code "creates" a pull request, and Bits Security Analyst "recommends" a verdict (as of September 2026; sources: Bits Investigation, Bits Code, Bits Security Analyst). The design leaves the final decision with a human. In operation, what may be delegated is "gather, line up, draft," and what humans keep is "adopt, merge, close." For irreversible decisions such as declaring an incident, treat Bits output as one piece of evidence and keep a record that a human decided. Even as Bits Remediation (Preview) extends into taking action, the line does not move. Before widening the scope, the order is to put per-user credit limits and the Audit Trail in place and build a setup in which someone capable of verifying the output reads it.
Things to watch out for
- This article is based on the official documentation as of September 2026. Bits Data Analysis, Bits Detection, and Bits Remediation (in part) carry a Preview marking as of that date, and their content and names may change at general availability.
- Because the unit prices of Bits AI and the MCP Server are not on any SKU row of the public pricing table, this article does not give numbers. Only the billing unit (AI Credits) and the names of the pricing tiers are stated, based on the official documentation. Check prices on the official pricing page.
- This article explains how billing works. For current unit prices and plans, refer to Datadog's official pricing page. This article makes no claims about the accuracy or performance of generative AI output.