Skip to main content

​
Deployment Guides

Docker Compose

Single VM, up to 100 developers

Kubernetes

Clustered, 100+ developers

​
Docker Compose Architecture

All services run as containers on a single Linux host, orchestrated by Docker Compose.
Greptile architecture overview

​
Core Services

​
Background Workers

​
Infrastructure Services

​
Data Flow

  1. Webhook received → greptile-webhook validates and queues the event
  2. Hatchet picks up the job and dispatches to appropriate worker
  3. Workers (chunker, summarizer, reviews) process via llmproxy
  4. Results stored in PostgreSQL, response posted back to SCM

​
Network Requirements

Must expose:
  • Port 3007 for SCM webhooks (or route through Caddy on 443)
  • Port 3000 for web UI access
  • Port 8080 for Hatchet admin (optional, can restrict to internal)
Must reach outbound:
  • LLM provider APIs (OpenAI, Anthropic, Bedrock, etc.)
  • SCM provider APIs (GitHub, GitLab, etc.)
  • Container registry for image pulls

​
Storage

PostgreSQL stores all application data including:
  • Repository metadata and summaries
  • Code embeddings (via pgvector)
  • Review history and analytics
  • User accounts and settings
Plan storage based on repository sizes. Embeddings are the largest component.

​
Kubernetes Architecture

Services deployed as pods across a Kubernetes cluster, managed by Helm charts. External PostgreSQL and Redis recommended for production.
Kubernetes architecture

​
Pod Deployments

Same services as Docker Compose, deployed as separate Kubernetes Deployments:

​
External Services

Unlike Docker Compose, Kubernetes deployments typically use managed services:

​
Networking

Ingress: LoadBalancer or Ingress controller exposes web and webhook services. Service mesh: Optional. mTLS between services if using Istio/Linkerd. Egress: NAT gateway for outbound traffic to LLM/SCM providers.

​
Scaling Considerations

  • API and Webhook scale with traffic volume
  • Chunker scales with new repository indexing load
  • Summarizer and Reviews scale with LLM throughput requirements
  • Use HPA (Horizontal Pod Autoscaler) for dynamic scaling based on CPU/memory

​
Security Model

​
Authentication

​
Secrets Management

Docker Compose: Environment variables in .env file. For production, use a secrets manager and inject at runtime. Kubernetes: External Secrets Operator syncing from AWS Secrets Manager, Vault, or similar.

​
Network Security

  • Deploy in private subnet, expose only webhook port externally
  • Database and Redis should not have public IPs
  • Use security groups/firewall rules to restrict access
  • All external traffic over TLS

​
Monitoring

​
Key Metrics

  • Logs: CloudWatch, ELK, or Loki
  • Metrics: Prometheus + Grafana, or CloudWatch
  • Alerting: PagerDuty, Opsgenie, or native cloud alerting
Greptile’s Hatchet dashboard (port 8080) provides workflow-level visibility without additional setup.