Skip to content

Production & Enterprise Deployment Guide 🚀

Scaling CommitVigil to 1000+ users requires moving from a local monolithic setup to a distributed, high-availability architecture.

🏗️ Enterprise Architecture

For large-scale teams, we separate the Ingestion Layer (API) from the Processing Layer (Workers).

  1. API Layer: Stateless FastAPI pods behind a Load Balancer (AWS ALB / Nginx).
  2. Processing Layer: Distributed arq workers scaling horizontally based on queue depth.
  3. State Layer: Managed Redis (AWS ElastiCache) and PostgreSQL (AWS RDS).

☸️ Kubernetes Orchestration

We provide production-ready K8s manifests in infra/k8s/.

Key Scaling Features:

  • Replica Sets: Run multiple API instances to handle high traffic.
  • Auto-Scaling (HPA): Workers automatically scale from 2 to 20+ pods based on CPU usage or queue backlog.
  • Resource Constraints: Strict CPU/Memory limits to prevent noisy-neighbor issues on shared clusters.

Deployment Command:

kubectl apply -f infra/k8s/

🔄 Automated Org-Wide Ingestion

Managers don't need to manually ingest commits. Use GitHub Webhooks at the Organization level:

  1. Go to GitHub Organization Settings -> Webhooks.
  2. Add a new webhook:
    • Payload URL: https://api.commitvigil.yourcloud.com/api/v1/ingest/git
    • Content Type: application/json
    • Secret: Your fixed API_KEY_SECRET
    • Events: Select Pushes.

Result: Every commit from every repo in your organization is automatically processed for accountability.

🆔 Bulk Identity Sync (SSO)

Mapping 1000+ users via curl is inefficient. Implement a Mapping Sync Service:

  • Active Directory / Google Workspace: Write a small cronjob that polls your employee list and calls:
  • POST /api/v1/users/config/slack (Email -> Slack ID)
  • POST /api/v1/users/config/git (Email -> Git Email)
  • Automatic Discovery: CommitVigil can be configured to use the Slack users.lookupByEmail API to auto-map users on their first commitment.

📊 Monitoring & Observability

Enterprise deployment includes Prometheus + Grafana integration:

  • SLIs (Service Level Indicators):
  • background_jobs_pending: How many commitments are waiting for AI processing?
  • ai_token_usage_total: Cost tracking per team/department.
  • intervention_acceptance_rate: How often are managers acting on AI pings?

💰 FinOps: Optimizing LLM Costs

At 1000+ users, token costs can add up. CommitVigil helps you optimize:

  1. Provider Switching: Use Groq (Llama 3.3 70b) for high-speed, low-cost extraction, and OpenAI (GPT-4o) only for complex tone adaptation.
  2. Mock Mode for Non-Critical Repos: Use LLM_PROVIDER="mock" for internal experimental repositories to save budget.
  3. Caching: CommitVigil can be extended with a vector cache (like RedisVL) to reuse extracted tasks from similar commit messages.

[!IMPORTANT] Always deploy in a Zero-Trust environment using encrypted secrets and private network VPCs for the Database and Redis.