Ship Internal Tools with AI Superpowers
Lacquer is a open-source AI workflow engine that turns repeatable engineering tasks into reliable YAML workflows that never skip a step. Think GitHub Actions, but for AI-powered internal tools.
inputs:
pod_name:
type: string
required: true
agents:
assistant:
provider: anthropic
model: claude-sonnet-4
system_prompt: |
You are a Kubernetes SRE expert. Analyze logs for: root causes, error patterns,
service impact, and specific remediation steps.
workflow:
steps:
- id: get_logs
run: "kubectl logs '${{ inputs.pod_name }}' --tail=50 | grep -E 'ERROR|WARN|Exception'"
- id: analyze_logs
agent: assistant
prompt: |
Analyze these recent error logs and identify root causes and recommended fixes:
${{ steps.get_logs.output }}
outputs:
issues: ${{ steps.analyze_logs.output }}
Stop Building Workflows No One Can Debug
Built for engineers who prefer terminals over drag-and-drop
GitOps Native
Your workflows are just YAML files. Commit them, review them, version them like any other code.
Zero Dependencies
No Python environments, no package conflicts, just a lightweight Go binary.
Local-First Development
Test everything on your laptop before deploying. No cloud account needed.
Familiar DSL
If you've used GitHub Actions, you'll feel right at home.
Declarative > Imperative
Describe what you want, not how to get it. Let Lacquer handle the orchestration complexity while you focus on business logic.
Production Ready
Built-in HTTP server, health checks, metrics, and observability. Deploy to Kubernetes, serverless, or regular VMs.
All the Features Engineering Teams Need
Everything you need to build production-ready AI workflows for internal tooling
steps:
- id: check_health
agent: monitor
prompt: "Check health status of service: ${{ inputs.service_name }}"
outputs:
healthy:
type: boolean
description: "Whether the service is healthy"
error_rate:
type: float
description: "The error rate of the service"
# Conditionally execute steps
- id: scale_up
condition: ${{ steps.check_health.outputs.error_rate > 0.05 }}
run: "kubectl scale deployment ${{ inputs.service_name }} --replicas=5"
# Break out steps into sub steps and run until a condition is met
- id: rolling_restart
while: ${{ steps.rolling_restart.iteration < 3 && !steps.rolling_restart.outputs.healthy }}
steps:
- id: restart_pod
run: |
kubectl rollout restart deployment/${{ inputs.service_name }}
kubectl rollout status deployment/${{ inputs.service_name }} --timeout=300s
- id: verify_health
agent: monitor
prompt: |
Verify service health after restart:
- Check HTTP endpoints return 200
- Verify error rate < 1%
- Confirm all pods are ready
Service: ${{ inputs.service_name }}
outputs:
healthy:
type: boolean
description: "Whether the service is healthy"
metrics:
type: object
description: "The metrics of the service"
agents:
incident_responder:
provider: anthropic
model: claude-sonnet-4
system_prompt: |
You are an SRE expert who:
- Analyzes production incidents
- Identifies root causes from logs and metrics
- Creates runbooks for remediation
- Documents post-mortems
tools:
- name: filesystem
description: Access runbooks and configuration files
mcp_server:
type: local
command: npx
args:
- "-y"
- "@modelcontextprotocol/server-filesystem"
- "/etc/kubernetes/manifests"
requirements:
runtimes:
- name: python
version: "3.9"
workflow:
steps:
- id: create_fix
agent: fixer
prompt: |
We've encountered the following error in production
${{ inputs.error }}
Please create a fix for the error in the following code:
${{ inputs.code }}
outputs:
patch:
type: string
description: The patch to apply to the code to fix the error
# Use 'run' steps when you need to execute custom logic that goes beyond
# simple agent interactions. These are bash scripts that are executed
# directly on the host system.
- id: validate_fix
run: "python3 scripts/validate.py"
with:
patch: ${{ steps.create_fix.outputs.patch }}
code: ${{ inputs.code }}
# Or use `container` steps when you want to execute custom logic in a more
# isolated environment. This is useful when you have complex dependencies.
- id: validate_fix_container
container: ./validate/Dockerfile
command:
- scripts/validate.py
- ${{ steps.create_fix.outputs.patch }}
- ${{ inputs.code }}
state:
rollback_count: 0
deployment_status: "pending"
workflow:
steps:
- id: deploy_service
run: "helm upgrade --install ${{ inputs.service }} ./charts/${{ inputs.service }}"
updates:
deployment_status: "${{ steps.deploy_service.output ? 'deployed' : 'failed' }}"
- id: rollback_if_needed
condition: ${{ state.deployment_status == 'failed' }}
run: "helm rollback ${{ inputs.service }}"
updates:
rollback_count: "${{ state.rollback_count + 1 }}"
agents:
ops_assistant:
provider: openai
model: gpt-4
temperature: 0.2
system_prompt: You investigate production issues and query infrastructure state.
tools:
- name: query_metrics
script: "python ./tools/prometheus_query.py"
description: "Query Prometheus for system metrics"
parameters:
type: object
properties:
query:
type: string
description: "PromQL query to execute"
timerange:
type: string
description: "Time range (e.g., '5m', '1h', '24h')"
Get Started in 60 Seconds
From install to first workflow in under a minute
Install
curl -sSL https://lacquer.ai/install.sh | sh
Single binary, zero dependencies
Create
laq init
Get AI to scaffold your first workflow
Run
laq run workflow.laq.yml
Execute and see the magic