Taskr

A self-hosted, production-grade background job processing platform — like a mini Trigger.dev or Inngest — that can run AI agent workflows.

Core Engine (Node + Express + BullMQ + Redis)

• REST API to enqueue jobs — with priority, delay, cron scheduling
• Worker pool that pulls and executes jobs concurrently
• Retry with exponential backoff + dead letter queue
• Job status tracking in PostgreSQL (with full audit log)
• Real-time job status via WebSockets (SSE fallback)

Multi-Tenancy Layer

• API key-based auth per tenant
• Per-tenant rate limiting (Redis-backed)
• Tenant isolation in PostgreSQL (Row-Level Security)

AI Agentic Layer (yahan tera differentiator hai)

• Ek job type: ai_agent_task — runs a RAG pipeline step or - tool-calling agent
• Multi-step agent workflows as chained jobs
• Output stored in MongoDB (because agent outputs are unstructured)

Infra

• Terraform: EKS cluster, RDS PostgreSQL, ElastiCache Redis on AWS
• K8s manifests: Deployment, HPA (auto-scale workers on queue - depth), PodDisruptionBudget
• GitHub Actions CI/CD: test → build Docker image → push to ECR → deploy to K8s
• OpenTelemetry tracing + Prometheus metrics + Grafana dashboard