AgentFlow allows engineering teams to construct resilient, self-healing LLM workflows. Built on a Directed Acyclic Graph (DAG) state engine, it supports tool calling, parallel worker execution, human-in-the-loop approval checkpoints, and memory persistence.
Back to All Projectsprojects/agentflow-orchestrator
AI & Systems
Featured Production Architecture
1,200+ Stars · 250+ Production Deployments
AgentFlow: Distributed Autonomous Agent Framework
An open-source orchestration runtime for multi-agent workflows with sandboxed tool execution and deterministic state graphs.
Key Architectural & Engineering Highlights
Sandboxed Python & Bash tool execution engine with memory limits
Visual DAG execution debugger built with SVG and React Flow
Zero-dependency lightweight client with sub-10ms serialization overhead
System Design & Architectural Deep Dive
Technologies Used:
TypeScript
Next.js
Node.js
pgvector
Docker
More Engineering Projects
Explore more architectures and open-source systems.
Open Source
50k+ Docker Pulls · CI IntegrationSchemaForge: Zero-Downtime DB Migration Engine
Developer CLI and validation pipeline preventing locking table locks during Postgres database migrations.
Go, PostgreSQL, GitHub ActionsCase Study
Web App
450ms → 12ms p95 Latency ReductionTurboStash: Intelligent Edge Cache Middleware
Smart tier-2 caching proxy for Next.js App Router and serverless backends with programmatic stale-while-revalidate invalidation.
TypeScript, Redis, Cloudflare WorkersCase Study