Quick Start

Getting the first agent running has two parts. First, you connect Autoheal to your code repositories, CI/CD pipelines, observability tools, cloud runtimes, issue trackers and coding agents. The factory then builds the context graph every agent works from. Second, you pick the post-coding workflow with the most manual toil, usually a customer-impacting one like incident response or vulnerability remediation. You deploy a pre-built agent to automate it, so you don't configure an agent from scratch.

With integrations in place, the first agent runs in minutes and can be deployed to a team in hours. Broader rollout follows your enterprise's access and deployment requirements. This guide uses the Incident Response agent as the first agent. It starts smaller than a full rollout, with one observability integration and one incident. It leaves authoring your own context until you see the agent work against your environment.

Prerequisites

Before you start, ensure you have:

• An Autoheal account (contact support@autoheal.ai for access)
• Project Admin on the project you're configuring, to connect integrations (see Roles & Permissions)
• Credentials for at least one observability tool (Datadog, Grafana, etc.)

Step 1: Log In to Autoheal

1

Access the Dashboard

Navigate to your Autoheal instance URL (e.g., https://<tenant>.autoheal.ai) and click Log In.

2

Authenticate

Sign in using your Email / Password or your organization's SSO provider.

Step 2: Connect Your First Integration

The Incident Response agent gathers its evidence from your integrations, so it needs at least one before it can run. Start with your observability tool.

1

Navigate to Integrations

From the sidebar, click Integrations.

2

Choose an Integration

Select the integration you want to connect. We recommend starting with your primary monitoring tool (e.g., Datadog, Grafana).

3

Enter Credentials

Provide the required API credentials. See Integrations for detailed instructions for each integration.

tip

An observability tool gives the Incident Response agent the metrics, logs and traces it needs for most incidents. Add source control next, so the agent can correlate an incident with recent deployments.

Step 3: Run the Incident Response Agent

The Incident Response agent is pre-built, so it works against your connected integration with no configuration. It runs a Deep investigation by default. It forms several hypotheses, checks each against the evidence, and returns a ranked root cause with a confidence score.

1

Open the Investigations Panel

From the sidebar, click Investigations.

2

Describe the Incident

Type a description of the incident you want the agent to work on. For example:

High latency alerts firing on the checkout service.
Started around 2:30 PM UTC.

3

Let the Agent Work

The Incident Response agent will:

• Query your connected integrations for relevant data
• Search your Engineering Context Graph for related skills and memories
• Analyze patterns and correlations
• Present a ranked root cause and proposed mitigations, with the evidence behind its reasoning

4

Follow Up

Ask follow-up questions to investigate further:

Can you check the database metrics during that time?

Are there any similar incidents in the past week?

Once this works on a manual run, connect PagerDuty or your alerting tool so the Incident Response agent starts automatically when an incident is declared.

Step 4: Add Your Own Context

After a few runs, Autoheal starts building context on its own. It captures memories from finished runs. The Evaluator scores each run. The Healer proposes skill and memory changes for you to review.

You can accelerate that by authoring context yourself. When the agent asks "which dashboard should I check?" or "who owns this service?", that's a gap worth filling.

1

Open Skills

From the sidebar, open Skills in the Library group.

2

Create a Skill

Click New skill, give it a title, and click Create. It opens in the agent workspace.

3

Add Content

Document a common incident type your team handles. Include:

• Symptoms and how to identify the issue
• Step-by-step remediation procedures
• Escalation paths if needed

4

Commit

Commit the skill from the workspace. It's then available to every agent allowed to use it.

Example Incident Response Run

Here's what a typical Incident Response run looks like:

User Query

"We're seeing 5xx errors spike on the payment service. Can you investigate?"

Agent Response

The Incident Response agent will:

1. Query Datadog/Grafana for error rates and latency metrics
2. Pull recent logs from the payment service
3. Check for recent deployments or changes
4. Search the Engineering Context Graph for payment service skills and memories
5. Present a ranked root cause with a confidence score and proposed mitigations

Follow-up Questions

You can then ask:

• "What changed in the last deployment?"
• "Show me the database connection pool metrics"
• "Are other services affected?"

Next Steps

Incident Response

How the agent investigates, mitigates, and surfaces proactive actions after an incident.

Connect More Integrations

Add alerting, source control and logging so the agent has more evidence to work from.

Author Your Skills

Write the alert and integration skills your agents follow.

Try Another Agent

Run Vulnerability Remediation, or build your own agent for any other post-coding step.

Invite Team Members

Add your team and configure permissions.

Getting Help

note

Need assistance? Contact our support team at support@autoheal.ai or check the detailed guides in this documentation.