> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rootly.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Atlas
> Atlas is Rootly's AI layer behind AI SRE, Slack conversations, web chat, mobile chat, and the Chat API. Learn where its context and controls live.
**Atlas is Rootly's AI layer.** It brings together the AI capabilities that answer operational questions, investigate incidents, and use the context and tools available to your organization. [AI SRE](/ai/ai-sre/overview) is one way to use Atlas: it runs a structured investigation of an alert or incident. You can also interact with Rootly AI by mentioning `@Rootly` in Slack, asking a question in the web or mobile chat, or sending a request through the Chat API.
Atlas is an umbrella name, not a single switch in the Rootly app. Its entry points and supporting context have separate controls. What each one can read or do depends on its surface, enablement, your permissions, and the tools connected to your team.
Where the consolidated AI SRE navigation is available, the sidebar has one **AI SRE** item with two tabs: **Investigations** lists the investigations AI SRE has run, and **Atlas** groups the controls behind them — entry points, context and tools, investigation settings, and organization-wide AI settings. This guide uses those paths.
The consolidated **AI SRE** navigation is rolling out and may not be enabled for your team yet. Until it is, the same settings live under **AI & Agents**, with the AI SRE settings under **AI & Agents → AI SRE**, and investigations under **Investigations** in the sidebar. See [AI Settings](/ai/ai-settings#where-ai-settings-live).
AI SRE has its own enablement path through your Rootly account team; turning on conversational Rootly AI does not enable investigations. Private Agent is an Early Preview for approved customers; access, including from Slack, also requires AI SRE to be enabled. The Knowledge graph and Memory can have separate availability requirements. The [Chat API](/api-reference/overview#chat-api) has its own team-level controls and also requires **Rootly Agent in Slack** to be enabled. See [AI Settings](/ai/ai-settings) and [Manage User Permissions](/managing-users/user-permissions#ai-agents-and-ai-sre).
## Ways to Use Atlas
| Entry point | What it does | Learn more |
| --------------- | ------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------- |
| **AI SRE** | Investigates alerts and incidents, tests possible causes against evidence, and records a result | [AI SRE](/ai/ai-sre/overview) |
| **Slack** | Responds to `@Rootly` in an incident channel or thread, as well as supported assistant and DM conversations | [Rootly AI in Slack](/ai/rootly-in-slack/overview) |
| **Web chat** | Answers questions from the incident page using incident context and its available read-only tools | [Rootly AI in Web](/ai/ask-rootly-ai) |
| **Mobile chat** | Answers questions in the mobile app and supports permitted incident actions with confirmation | [Rootly AI on Mobile](/ai/rootly-ai-on-mobile) |
| **Chat API** | Lets an application send a chat message, optionally bound to an incident or alert, and receive a reply or stream it | [Chat API endpoints and availability](/api-reference/overview#chat-api) |
The same context source is not necessarily available on every entry point. For example, the Knowledge graph and Memory described below are currently used by AI SRE investigations; web chat has its own tool allowlist. Use each surface's guide for its behavior and permissions.
## Context and Tools
| Part | What It Is | Where to Find It | Current Use |
| -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
| [AI Connectors](/ai/connectors/overview) | Live access to your SaaS observability, code, cloud, work-tracking and knowledge tools | **AI SRE → Atlas → Connectors** (**AI & Agents → Connectors** if your sidebar doesn't have an **AI SRE** item) | AI SRE investigations and supported conversational surfaces |
| [Private Agents](/private-agent) | An Early Preview for approved customers: an outbound-only agent you run inside your network that exposes typed capabilities for systems such as Kubernetes, Prometheus, databases and internal HTTP services | **AI SRE → Atlas → Private agents** | AI SRE investigations and the Rootly Agent in Slack |
| [Knowledge Graph](/ai/atlas/knowledge-graph) | Derived facts about your services, infrastructure and how they relate, stored with source and internal confidence; the page shows source and evidence labels rather than confidence scores | **AI SRE → Atlas → Knowledge graph** | AI SRE investigations, and **Generate with AI** on the Instructions page |
| [Memory](/ai/atlas/memory) | Short, durable learnings about your systems that AI SRE proposes and your team curates | **AI SRE → Atlas → Memory** | AI SRE investigations |
The **Knowledge graph** and **Memory** cards sit beside **Instructions** and **Investigation rules**. They are documented here because they are part of the AI layer, even though their current reader is AI SRE. Instructions and Investigation rules configure investigations.
## How AI SRE Uses Atlas
[AI connectors](/ai/connectors/overview) give AI SRE live access to the SaaS tools your team already uses. [Private agents](/private-agent) reach systems inside your network over an outbound-only connection, so you don't expose them publicly.
Rootly reads your Rootly services and catalog, plus supported connectors such as AWS, Google Cloud, GitHub, Datadog and Grafana. It stores what it finds as facts: what exists, what depends on what, and who owns it. Updates run in the background, and disconnecting a connector removes the facts it contributed.
AI SRE matches the names in an alert or incident to entities in the graph. It walks their dependencies to find candidate causes and map what else could be affected, recalls the best-matching Active notes from Memory, and queries your connectors and private agents for live evidence. It attributes a cause only when evidence gathered during the investigation supports it.
After an investigation, AI SRE can propose notes to [Memory](/ai/atlas/memory). AI SRE doesn't recall a proposed note until it's Active. An owner or admin selects **Activate**, or, where automatic activation is enabled, an eligible note activates on its own after proving helpful in at least two separate investigations. AI SRE also quarantines notes on its own, for example when live evidence contradicts one.
AI connectors and private agents keep your providers' permission boundaries, and they aren't guaranteed to be read-only. Built-in connector tools can accept provider-defined commands or queries that change data. A Custom MCP connection exposes whichever tools you check, and a private agent can offer write-capable HTTP and SQL capabilities when its local policy allows them. Responder-started investigations use the initiating user's Private Agent permissions; automatic investigations use the `ai-sre` system actor with no responder in the loop. Expose only the tools, local policies and credentials that are safe for unattended use, and prefer read-only identities. See [Evidence Sources](/ai/ai-sre/evidence-sources) for each boundary.
## AI SRE Settings
Two settings in the AI SRE section decide how and when investigations run.
| Setting | What It Controls | Where to Find It |
| ----------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------- |
| [Instructions](/ai/ai-sre/instructions) | How AI SRE investigates: team-wide guidance in plain language, such as evidence to prioritize or systems to check. AI SRE follows instructions but doesn't treat them as evidence. | **AI SRE → Atlas → Instructions** (**AI & Agents → AI SRE → Instructions** if your sidebar doesn't have an **AI SRE** item) |
| [Investigation Rules](/ai/ai-sre/investigation-rules) | When AI SRE investigates an alert: each rule matches alerts and sets a **Run mode** of **Manual**, **Auto-run** or **Paused**, optionally with its own instructions. | **AI SRE → Atlas → Investigation rules** |
When an investigation misses a fact about your environment, fix the source it comes from: a connector, a Rootly service or a catalog field. When AI SRE should work differently, such as checking recent deploys first, write that in Instructions or in a rule's instructions.
Where it's enabled, **Generate with AI** on the Instructions page drafts instructions from your Knowledge graph facts, some of your connectors and your investigation rules.
## Improve AI SRE Investigations
1. **Connect your primary observability provider, your code or deployment provider, and the knowledge source that holds your runbooks.** Add other connectors when an investigation shows a specific gap. If you connect Datadog with OAuth, change and deploy correlation isn't included, so use API and application keys if AI SRE needs to investigate what changed.
2. **Fill in your Rootly services and [catalog](/catalogs).** A service's Datadog ID, GitHub repository, PagerDuty ID and Kubernetes deployment name let AI SRE match an alert to that service. Its dependencies, owners and tier fields feed the Knowledge graph. When sources disagree about what a thing is, declared data like this outranks anything inferred from telemetry.
3. **Describe your infrastructure.** Connect AWS or Google Cloud, or give the GitHub connector access to your infrastructure-as-code repository. Rootly picks that repository automatically. See [Knowledge Graph](/ai/atlas/knowledge-graph) for every source it reads.
4. **If approved for the Private Agent Early Preview, add a [private agent](/private-agent) for systems the internet can't reach**, such as an in-cluster Prometheus or an internal database.
5. **Review [Memory](/ai/atlas/memory).** Activate useful proposed notes, and deprecate or quarantine notes that are obsolete or misleading. Only Active notes reach later investigations. Manual status changes require an Incident Response Owner or Admin with a seat; where enabled, AI SRE can also activate, deprecate, quarantine, or replace notes automatically under the conditions in the [Memory guide](/ai/atlas/memory#what-ai-sre-changes-on-its-own).
Let completed investigations show you what's missing. A run that ends **Could not investigate** stopped before it gathered enough evidence, so check that the connectors for that alert's source are connected and authorized. For **Inconclusive — needs human**, look in the **Investigation path** for checks that were inconclusive or gathered no evidence, then read **Recommended next steps**. See [Reading the Result](/ai/ai-sre/reading-the-result).
## Related Pages
Explore the services, infrastructure and relationships AI SRE uses, with the evidence behind each relationship.
Connect observability, code, cloud, work-tracking and knowledge tools under **AI SRE → Atlas → Connectors** (**AI & Agents → Connectors** if your sidebar doesn't have an **AI SRE** item).
Review, activate, deprecate or quarantine the learnings AI SRE recalls in later investigations.
Reach systems inside your network through an outbound-only, policy-controlled agent.
Investigate alerts and incidents through one of Atlas's entry points.
See what an investigation reads from each part of Atlas, and the access boundaries that apply.
Ask Rootly AI questions and take permitted actions in Slack.
Ask questions about an incident from its web page.
## Frequently Asked Questions
Atlas has no switch of its own. The **Atlas** tab gathers the controls, but each one is enabled separately: configure Slack, web, and mobile chat under **AI SRE → Atlas → Conversations** (**AI & Agents → Features** if your sidebar doesn't have an **AI SRE** item), connections under **AI SRE → Atlas → Connectors** or **Private agents**, and investigations under **AI SRE → Atlas → Instructions** and **Investigation rules**. The Chat API has separate team-level controls and also requires the **Rootly Agent in Slack** toggle. Your Rootly account team enables AI SRE separately; turning on **Opt in to Rootly AI capabilities** doesn't enable investigations.
Atlas is the AI layer behind the entry points above, but they do not all use the same tools or data. Currently, AI SRE reads the Knowledge graph and Memory; connectors also serve supported conversations, and web chat uses a narrower read-only tool set. Check the relevant surface guide before assuming a capability is available there.
Yes. Open **AI SRE → Atlas → Knowledge graph**, search for the service and select it. The detail panel lists its attributes and each relationship with its evidence and sources, such as **Observed · Datadog**. The page is view-only, so correct a fact in the system it came from. To see what AI SRE has learned from investigations, open **AI SRE → Atlas → Memory**.
For what a thing is, such as its kind, tier or environment, declared sources like infrastructure as code, the catalog and your cloud accounts outrank telemetry, and confidence breaks ties. Relationships from every source are kept and labeled by evidence: **Observed**, **Declared**, **Capability** or **Corroborated**. When Rootly finds an environment or resource type conflict it can't settle, the Knowledge graph page shows **Some graph data needs attention**.
No. AI SRE recalls only Active notes, treats each one as a hypothesis to check against live evidence, and can't conclude a root cause from a note alone.
Rootly deletes the facts that connector contributed when you disconnect it. Facts from your Rootly services and catalog stay.