Getting started

Aviator Agents is currently supported only in an on-premise environment. If you are interested in trying out our beta cloud hosted option, please email us at [email protected].

Onboarding

This setup guide will walk you through the initial set up for Agents workflow. Before getting started, make sure you have installed Aviator Agents on your private cloud. The guide below assumes you have setup Aviator on the domain https://aviator.example.com.

1. Create an account: https://aviator.example.com/auth/login

2. Follow the onboarding flow to connect to the Aviator GitHub app, authorize one or more repositories that you want to use with Agents. You can always add more repositories later.

If you have trouble connecting the app, please read the troubleshooting doc.

Workflow

Aviator Agents workflow is divided into 3 phases:

• Planning

• Execution

• Review

Planning

This is the phase where you work with the agents to plan out a work.

1. Once a repository is connected, go to the Agents dashboard: https://aviator.example.com/agents/dashboard

2. On the chat prompt, describe briefly the task that you want to collaborate with the agents on, and select the repository to start with.

Start a new plan

1. This will create a new chat session, where the agents will build a Runbook after analyzing the code. Building the Runbook is a multistep process. During this process, the agents will:

   1. Fetch the code from GitHub

   2. Inspect the query

   3. Analyze the code base to build the context

   4. Understand dependencies

   5. Ask clarifying questions

2. You can provide feedback via the chat interface or modify the Runbook manually. Runbooks follow a specific pattern and the agents will validate it on every edit.

Execution

1. To run one or more steps in the Runbook - click on Review on the right pane, and select the step you want the agent to try and click “Run”. You can also dry run a specific step for a specific file by clicking on the menu and following the dry run prompt.

2. As agents start running in background, they will post some verbose logs in the logs viewer. Once completed, agents will create one or more PR and share the links within the corresponding Runbook step.

3. You can also ask agents to run the whole Runbook autonomously, where agents will run each step in-order and create relevant PRs.

Review

1. For any PRs generated by the agents, you can review the changes directly in GitHub similar to how you review a PR created by a developer.

2. After a review is submitted, the agents will analyze all the comments and make corresponding edits.

3. You can go back and forth with the agents requesting multiple iterations to improve the code. Once you are satisfied with the changes, you can merge the PR manually or using MergeQueue.