Skip to content

About

Build Your Own Herness with SoloEngine. Achieve Loop Engineering — by simply dragging and dropping.

Topics

Resources

Code of conduct

Contributing

Stars

36 stars

Watchers

8 watching

Forks

Latest commit

 

History

233 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SoloEngine

Build Your Custom Harness in 5 Minutes.


SoloEngine is the first open-source low-code platform purpose-built for crafting Harness. Drag an Agent onto the canvas, wire up its collaborators, and hit run — it will autonomously plan, act, observe, and iterate until the task is done.

No Workflow. No orchestration code. Just Agents that get things done.

Stars Node pnpm TypeScript React License Ask Zread

English | 简体中文 | Español | Deutsch | Français


Why SoloEngine

Agentic AI and Harness Engineering are reshaping the software landscape — a single developer can now deliver what once took a ten-person team. Yet this revolution has remained confined to the IDE. To build a Harness that truly fits your business, you have to hand-write LangChain pipelines, repeatedly debug ReAct loops, and define tool Schemas one by one. Don't know how to code? Then none of this is within reach.

The alternatives don't help: workflow platforms (Dify, n8n, etc.) run pre-scripted paths, not autonomous agents. Code frameworks (LangChain, CrewAI, etc.) require Python. SoloEngine fills the gap.


Dify, n8n, Zapier LangChain, CrewAI, LangGraph SoloEngine
Agentic AI ✗ Scripted workflows only ✓ ReAct / multi-Agent ✓ ReAct / multi-Agent
No code required ✓ ✗ Python mandatory ✓
Visual composition ✓ Full canvas ✗ ✓ Full canvas
Domain expert can build alone ✓ ✗ ✓
Multi-Agent collaboration ✗ ✓ ✓
  • Kill WorkFlow — SoloEngine is fully ReAct-architecture-based. No preset workflows — just Think → Act → Observe → Repeat loops. Every decision is made dynamically at runtime. When an Agent hits an unexpected roadblock, it re-plans.
  • Domain experts build directly — A lawyer drags a Contract Review Agent onto the canvas, wires it to a Research Agent, hits run. No programmer in the loop.
  • Tools, Skills, MCP — hot-pluggable design, progressive disclosure — Each Agent loads only the tools and Skills it declared. Thanks to progressive disclosure, token consumption drops by over 85%.
  • Model-agnostic — Model access is a row of configuration (name, model, base URL, key) that becomes a runtime route. The adapter speaks the OpenAI and Anthropic protocols through pi-ai, so DeepSeek, OpenAI, Anthropic, Ollama, and OpenAI-compatible gateways share one interface.

How It Works

All Agents share the same underlying architecture — just configured differently. The canvas is compiled into executable Agentic AI.

  1. Compilation — Compiling is a phase inside a run, not a separate build step: the canvas is topologically sorted into an Agent DAG when a run starts. One compiler, infinite teams.
  2. One ReAct engine — Every Agent runs the same Think → Act → Observe → Repeat loop.
  3. Progressive disclosure — Agents only load what they use at runtime, making low-code agentic AI practical.
SoloEngine visual canvas demo

Compose Agents on the visual canvas and compile the team configuration into an executable Agent DAG

Quick Start

git clone https://github.com/Sh4r1ock/SoloEngine.git
cd SoloEngine

# Dependencies (Node >=24.9, pnpm)
pnpm install

# Self-check, then start the single-port runtime on 8991
node soloengine-launch/bin.mjs doctor
node soloengine-launch/bin.mjs start

The launcher prints the ready address, including the one-time process token that a fresh browser exchanges for the signed session cookie. The console and the run panel share that single port.

Sign in with the seeded superuser (system / system), configure a model under Model, then build your first Agent team.

Use Cases

  • VibeLawing — A lawyer drags search, filing, and formatting Agents onto the canvas in sequence, then hits run. AI will automatically break down and plan the legal work: parsing facts, locating statutes, organizing cases, structuring arguments, and formatting documents — all you need to do is review and fine-tune before delivery. The whole process feels as natural as a developer vibe-coding in Cursor.
  • VibeMarketing — A marketer drags research, copywriting, and asset Agents onto the canvas, then hits run. SoloEngine will automatically analyze the audience, then research relevant product approaches based on the findings, and benchmark against competitors to craft marketing strategies — delivering a polished marketing plan to you.
  • One-click package — Build your Agent team, hit package, one product ready for anyone.
SoloEngine run demo

Observe Agent planning, actions, tool calls, and iteration in real time from the run panel

Star History

Star History Chart

⭐ If you find this useful, give us a star — it helps a lot!

Acknowledgments

Special thanks to:

MiMo

Contributing

We are so excited to meet you.

Every typo fix, new tool plugin, doc tweak, or full-on feature lands — all of it makes SoloEngine better. Big PRs and small ones, both welcome.

📝 Contributing guide  ·  🐛 Good first issues  ·  📧 sh4r1ock@qq.com

License

Apache License 2.0. See LICENSE.


SoloEngine — AI for Every Industry ❤️

About

Build Your Own Herness with SoloEngine. Achieve Loop Engineering — by simply dragging and dropping.

Topics

Resources

Code of conduct

Contributing

Stars

36 stars

Watchers

8 watching

Forks

Releases

Packages

Contributors

Languages