Skip to main content

AI Assistant

Last updated on
New in 3.0

The AI Assistant has been reimagined with a fundamentally new architecture. It is now integrated directly into the navigation, available on all pages, and capable of both answering questions and performing actions. Access it via the lightning bolt icon in the top-right header.

AspectBefore (NG)After (3.0)
PlacementFloating button overlayIntegrated in navigation header
AvailabilityLimited to specific pagesAvailable on all pages
CapabilitiesAnswer-onlyQuestions and actions
AccessFloating button, easy to missLightning bolt icon in top-right header, always visible

Architecture​

The AI Assistant is composed of two systems that work together: Unified Chat for conversational queries and the Harness DevOps Agent for performing actions. Each system uses a different model optimized for its purpose.

Handles conversational queries, read operations, and knowledge retrieval. Routes requests to the appropriate backend integration.

  • Model: Sonnet 4.0

Capabilities​

The AI Assistant currently focuses on pipeline workflows with integrations into the Harness Knowledge Graph and Dashboard Agent. It supports all read operations and partial write operations through the DevOps Agent.

CapabilityStatusDescription
Pipeline OperationsSupportedCreate new pipelines and update existing pipeline configurations via the DevOps Agent. This is the primary write operation currently available.
Read Operation QueriesSupportedQuery any Harness resource — list pipelines, inspect execution details, view connector configurations, check service definitions, and more.
Knowledge GraphIntegratedConnects to the Harness Knowledge Graph to answer questions about platform concepts, configuration syntax, best practices, and troubleshooting guidance.
Dashboard AgentIntegratedRetrieves analytics and metrics from Harness dashboards. Can also create new dashboards, create widgets within dashboards, and answer questions about build trends, deployment frequency, failure rates, and other operational data.
Other Resource MutationsComing SoonCreate/update operations for secrets, services, environments, connectors, and other Harness resources.
Current Scope

The AI Assistant currently works with Pipelines as the primary resource for write operations. All other DevOps Agent actions (read queries, knowledge graph lookups, dashboard analytics) are fully operational.

Panel Layout​

The AI Assistant opens as a panel on the right side of the viewport, occupying approximately 30–40% of the screen width. The panel contains an input field at the bottom for typing questions and a scrollable content area above for viewing responses. The main page content remains visible and interactive alongside the panel.

AI-Powered Actions​

The AI Assistant can perform actions through two channels: Unified Chat (Sonnet 4.0) handles read queries and knowledge lookups, while the DevOps Agent (Opus 4.5) handles pipeline mutations.

ActionChannelDetails
Create PipelinesDevOps AgentGenerate new v1 pipeline YAML from natural language descriptions. The agent produces valid pipeline configurations using the v1 schema.
Update PipelinesDevOps AgentModify existing pipeline configurations — add stages, update steps, change triggers, adjust inputs and environment variables.
Read QueriesUnified ChatQuery any Harness resource: list pipelines, view execution history, inspect connectors, check service definitions, review environment configurations, and more.
Knowledge Graph LookupUnified ChatSearch Harness platform knowledge for concepts, configuration syntax, expression references, best practices, and troubleshooting guidance.
Dashboard AnalyticsUnified ChatQuery the Dashboard Agent for build metrics, deployment frequency, failure rates, DORA metrics, and other operational analytics. Can also create new dashboards and add widgets to existing dashboards.
Action Confirmation

All mutating actions (create, update) require explicit confirmation before execution. The DevOps Agent presents the proposed pipeline YAML and asks for approval before applying changes to your project.

Example Interactions​

The following examples demonstrate common interaction patterns with the AI Assistant.

Creating a Pipeline​

Ask the AI Assistant to create a new pipeline. The DevOps Agent (Opus 4.5) generates valid v1 pipeline YAML from your description.

User: Create a CI pipeline that builds a Docker image and pushes
      it to ECR when code is pushed to the main branch.

AI:   I'll create a CI pipeline with the following configuration:
      - Trigger: push to main branch
      - Stage: Build and push Docker image to ECR
      - Uses your existing "aws-connector" for ECR access

      Here's the pipeline YAML:

      pipeline:
        stages:
          - name: build-and-push
            steps:
              - name: build-image
                run:
                  script: docker build -t my-app .
              - name: push-to-ecr
                run:
                  container: amazon/aws-cli
                  script: |
                    aws ecr get-login-password --region $REGION | \
                      docker login --username AWS --password-stdin $ECR_REGISTRY
                    docker push $ECR_REGISTRY/my-app:latest
                  env:
                    REGION: <+input>
                    ECR_REGISTRY: <+input>

      Shall I create this pipeline in the current project?

Updating a Pipeline​

Modify an existing pipeline through a conversational interface. The DevOps Agent reads the current YAML and applies your requested changes.

User: Update the "deploy-prod" pipeline to add an approval step
      before the production deployment stage.

AI:   I found the "deploy-prod" pipeline. I'll add an approval
      step before the "deploy-to-production" stage.

      Changes to pipeline YAML:
      + Added step "approval" before deploy-to-production:
          - name: approval
            run:
              script: echo "Waiting for approval..."
            # Approval gate configured via Harness UI
      + Updated stage ordering to ensure approval runs first

      Would you like me to apply this change?

Read Query and Troubleshooting​

Use read operations to inspect resources and diagnose failures. Unified Chat (Sonnet 4.0) queries the platform and provides analysis.

User: My "deploy-staging" pipeline failed on the last run.
      What went wrong?

AI:   I queried the execution history for "deploy-staging".
      Last run: #247 (failed).

      Root Cause: The "apply-manifests" step failed with:
        "Unable to connect to Kubernetes cluster: connection
         refused on 10.0.1.50:6443"

      This indicates the delegate cannot reach the Kubernetes
      API server. Possible causes:
      1. The cluster endpoint has changed
      2. Network policy is blocking delegate access
      3. The cluster is temporarily unavailable

      Recommended Actions:
      - Verify the Kubernetes connector endpoint is correct
      - Check delegate connectivity to 10.0.1.50:6443
      - Review network policies in the delegate namespace

      Would you like me to look up the connector configuration?

Chat Features​

The AI Assistant includes built-in features for managing conversations, providing feedback, and maintaining a searchable history of past interactions.

FeatureDescription
User FeedbackRate any AI response with thumbs up or thumbs down. Feedback is used to improve response quality over time.
Copy MessagesCopy any chat message to your clipboard. Useful for sharing AI-generated YAML, commands, or explanations with your team.
Chat HistoryAll conversations are saved automatically. Search past chats, rename them for easy reference, or delete conversations you no longer need.

Chat History Management​

ActionDescription
SearchSearch across all past conversations by keyword to quickly find previous answers, pipeline configurations, or troubleshooting sessions.
RenameRename older chats with descriptive titles for easier navigation and organization.
DeleteDelete individual conversations from your chat history when they are no longer needed.

Settings​

The AI Assistant can be configured through Settings to control agent behavior, enable personalization through memories, and manage conversation history. Click on the ... dropdown menu and select Settings.

Rules​

Rules govern how the AI Assistant behaves and responds. They are plain Markdown files that provide instructions, constraints, and context to the agent. Rules can be defined at multiple scope levels, with narrower scopes taking precedence.

ScopeDescription
AccountOrganization-wide rules that apply to all users across the entire Harness account. Set by account admins to enforce company-level standards.
OrganizationRules scoped to a specific organization within the account. Useful for team-level conventions and guidelines.
ProjectRules scoped to a specific project. Define project-specific context like naming conventions, deployment targets, or tech stack details.
PersonalUser-level rules that apply only to your own AI interactions. Customize response style, preferred formats, or personal workflow preferences.
Rule Format

Rules are written as plain Markdown files. You can include instructions like "Always use v1 pipeline schema", "Prefer Kubernetes deployments over ECS", or "Include rollback steps in all production pipelines". Rules at narrower scopes (Project > Org > Account) take precedence when there are conflicts.

To create a rule, click + New rule on the Rules tab.

Memories​

The AI Assistant can learn from your chat and session interactions through Memories. Memories capture key context from conversations such as your preferred tools, common configurations, and project patterns, and summarize them to personalize future interactions.

  • Automatic Summarization: Memories are automatically summarized from chat and session interactions to capture important context without storing full conversation logs.
  • Personalization: Stored memories are used to personalize future responses, reducing repetitive explanations and aligning suggestions with your workflow.
  • User Control: Memories can be enabled or disabled at any time. When disabled, the AI Assistant will not store or reference any memories from your interactions.

Navigate to the Memories tab and click the checkbox to Enable Memories to learn your preferences automatically from chats.

Module Availability​

The AI Assistant is available across all Harness modules. It provides context-aware assistance tailored to the module you are currently working in.

ModuleAI Available
Continuous Integration (CI)✓
Continuous Delivery (CD)✓
Pipelines✓
Dashboards✓