AgentFlow

AgentFlow is a fullstack monorepo for building, running, and observing agentic workflows in production.

What it does

• Compose and run agent workflows
• Execute flows via API/runtime services
• Monitor usage, execution logs, and operational metrics
• Connect business systems (OAuth/integrations)

Architecture

flowchart LR
    Operator["Operator"]
    Studio["React workflow studio"]
    Dashboard["React operations dashboard"]
    API["NestJS API and orchestrator"]
    Queue["BullMQ / Redis"]
    Worker["Python async worker"]
    Data["PostgreSQL"]
    Systems["LLMs and business systems"]
    Telemetry["OpenTelemetry"]
    Observability["Prometheus / Grafana"]

    Operator --> Studio
    Operator --> Dashboard
    Studio --> API
    Dashboard --> API
    API --> Queue
    Queue --> Worker
    API --> Data
    Worker --> Data
    Worker --> Systems
    API --> Telemetry
    Worker --> Telemetry
    Telemetry --> Observability

Loading

The API owns workflow definitions, execution state and integrations. Long-running work is delegated to asynchronous workers, while telemetry provides an operational view across services.

Monorepo structure

• api/: NestJS API (REST/WebSocket, integrations, orchestration)
• web/studio/: flow editor and operator UI
• web/dashboard/: metrics and analytics dashboard
• worker/: Python async/background task worker
• infra/: local/dev infrastructure definitions

Tech stack

• TypeScript, Node.js, pnpm workspaces
• NestJS, BullMQ, PostgreSQL/SQLite
• React + Vite (studio/dashboard)
• OpenTelemetry + Prometheus/Grafana
• Jest/Vitest/Playwright/Cypress testing

Production-oriented design

• Queue-backed execution separates API latency from long-running agent work
• Workflow payloads use a documented JSON schema
• Observability spans API, workers and infrastructure
• CI builds the services and runs unit and API end-to-end tests
• Infrastructure definitions support local development and deployment

Further reading:

• docs/rfc/agent_dsl_rfc.md — workflow DSL
• docs/flow-payload.schema.json — payload contract
• api/src/docs/architecture/worker.md — worker architecture
• docs/ui/design-system.md — interface system

Security and compliance notes

• Secrets are injected through environment variables.
• Terraform state, local kube configs, and private keys must never be versioned.
• All historical state artifacts were removed from this publicized history.

Quick start

pnpm install
pnpm --filter api run start:dev
pnpm --filter web/studio run dev
pnpm --filter web/dashboard run dev
python3 worker/agentflow_worker.py

Testing

pnpm --filter api run test
pnpm --filter web/studio run test
pnpm --filter web/dashboard run test

Status

Public technical showcase repository focused on architecture, reliability, and production-ready agent workflow patterns.