AI Agents Guide

Atlas Studio’s 7 agents share a common interaction model: you describe what you need in natural language, the agent plans a multi-step execution, shows you the plan for approval, executes it, and returns validated results.

Common Agent Controls

Control	Purpose
Prompt bar	Describe your intent in plain English
Model selector	Choose Fast / Balanced / Deep per prompt
Run / Stop	Execute or cancel the current agent task
History	Navigate previous sessions in this agent
Export	Save results as Markdown, JSON, or YAML
Chain to Workflow	Convert results to a workflow node

RCA Agent

Root cause analysis — from incident ID to root cause in minutes.

When to use

You have a PagerDuty / OpsGenie / ServiceNow incident and want to understand why
You need to write a post-mortem with timeline and evidence
You want to identify which service or change caused an alert storm

How to use

Connect an ITSM source

Ensure you have a ServiceNow, PagerDuty, OpsGenie, or Jira connector configured (Connectors → Add). See Connectors.

Open the RCA Agent

Click RCA Agent in the left sidebar (or Cmd+1).

Describe the incident


Analyze incident INC0012345 — high latency on payment-service since 14:30 UTC

Or just paste an alert summary and let the agent find related incidents.

Review the AI plan

The agent shows the steps it will take: fetch incident, pull correlated alerts, traverse dependency graph, check recent deployments. Approve to proceed.

Inspect the evidence panel

The agent returns:

Timeline — sequence of events with timestamps
Root Cause — ranked hypotheses with confidence scores
Evidence — linked alerts, logs, deployment events
Affected Services — dependency graph visualization
Recommended Actions — with links to runbook templates

Export or create a runbook

Click “Create Runbook” to hand off the root cause to the Runbook Agent.

Example prompts


Find root cause for the payment-service 503 errors that started at 2:30 PM
Why did the database connection pool exhaust? Trace back to the cause.
Generate post-mortem for incident INC-9821 with timeline and contributing factors

Runbook Agent

Convert incidents into executable, step-by-step runbooks with approval gates.

When to use

You want to document a known resolution procedure
You need an executable playbook for on-call engineers
You want to convert a runbook to a workflow for automation

How to use

Open the Runbook Agent

Click Runbook Agent (or Cmd+2).

Describe the incident or procedure


Create a runbook to restart the payment-service pod in Kubernetes when connection pool exhausts

Review the generated steps

The agent produces numbered steps with:

Action (e.g., “kubectl rollout restart deployment/payment-service”)
Risk level (Low / Medium / High)
Validation (how to verify the step succeeded)
Rollback (how to undo if it fails)

Add auth and CI targets

Specify which runner or execution target should execute each step.

Convert to workflow (optional)

Click “Convert to Workflow” — the runbook becomes a DAG workflow with approval gates at high-risk steps.

Workflow Agent

Build production-grade DAG workflows from natural language.

When to use

Automate a recurring operational task
Orchestrate multi-system processes (ETL, batch, file transfer)
Replace a legacy scheduled job with an AI-managed workflow

How to use

Open the Workflow Agent

Click Workflow Agent (or Cmd+3).

Describe your workflow


Build a workflow that runs every weekday at 9 AM, queries the sales database, 
generates a CSV report, and uploads it to S3 with PGP encryption

Review the generated DAG

The agent creates a visual DAG (directed acyclic graph) with nodes for each step. You can:

Drag nodes to reorder
Click a node to edit its configuration
Add manual approval gates
Set SLA deadlines

Configure node parameters

Each node type has specific configuration. The agent pre-fills based on your description:

SQL Query — data source, query, output format
File Transfer (MFT) — source, destination, PGP keys
SAP RFC — function module, parameters
Notification — Slack/email with dynamic content

Publish to Web Portal

Click Publish to save the workflow to your Web Portal where it can be scheduled and run by the runner fleet.

Supported node types

Category	Nodes
Scripts	Bash, Python, Node.js, SQL
File Transfer	SFTP upload/download, S3, Azure Blob
Enterprise	SAP RFC, SAP OData, SAP IDoc, Mainframe JCL
Kubernetes	kubectl exec, helm upgrade, pod restart
Cloud	AWS CLI, Azure CLI, GCP CLI
Messaging	Kafka publish/consume, RabbitMQ, IBM MQ
HTTP	REST API calls, webhooks
Flow	Condition, loop, parallel, wait, event trigger
Notifications	Slack, email, PagerDuty, Teams

Coding Agent

Write and execute scripts for ops and dev tasks. Free with BYOM.

When to use

Generate a bash/Python script to fix an infrastructure issue
Write SQL to query a production database
Debug a deployment configuration
Generate Terraform / Helm / Kubernetes YAML

How to use


Write a Python script to parse Nginx access logs and find the top 10 IPs by request count


Generate a Kubernetes CronJob YAML for running a backup every night at 2 AM


Debug this Dockerfile — it fails at the npm install step with an EACCES error

BYOM setup

The Coding Agent is free — no Atlas credits consumed when you use your own model API key.

Settings → AI → BYOM → enter your OpenAI / Anthropic API key
Select the Coding Agent → model auto-switches to BYOM

Dashboard Builder

Generate live operational dashboards from natural language.

How to use


Build a dashboard showing payment-service latency (p50/p95/p99), error rate, 
and pod count. Include a 24-hour trend.

The agent:

Queries your connected Prometheus/Datadog/Elasticsearch
Generates PromQL/queries automatically
Shows a live preview with real data
Allows editing widget by widget
Exports as Grafana JSON or saves to Web Portal

Supported data sources

Prometheus / Thanos
Datadog
Elasticsearch / OpenSearch
CloudWatch
Any OpenTelemetry-compatible backend

Report Generator

Generate executive and operational reports.

Report types

Type	Description
SLA Report	Incident count, MTTR, SLA compliance by service
Incident Summary	Post-mortem narrative with root causes
Cost Analysis	Cloud spend by service, team, or feature
Compliance Bundle	SOC2/HIPAA/GDPR evidence for auditors
Custom	Any report from your data sources

How to use


Generate a weekly SLA report for the payments team covering this week. 
Include top 5 incidents, MTTR trend, and on-call hours.

Export as Markdown (for Confluence/Notion), PDF, or JSON (for downstream systems).

Data Analyzer

Query any connected data source in natural language.

How to use


Show me all orders placed in the last 24 hours that have status = failed

The agent translates to SQL / query DSL, runs it against your connected data source, and displays results as a table and bar chart.

Multi-turn refinement:


Now group by customer region
Filter to only US customers
Export this as CSV

Chaining

After analysis, click “Build Dashboard” or “Generate Report” to pass results to other agents.