@opcpflow/engine

OpcpFlow

Open DAG Workflow Framework — AI agent assembly, multi-type asset composition, and D4 self-evolving TypeScript DAG framework.

Includes a visual editor, 11 AI orchestration node types, Ready Frontier event-driven engine, ContextStore three-level memory, and D4EvolutionHook for automatic workflow evolution.

Why OpcpFlow?

• D4 Self-Evolution — Workflows evolve from static to dynamic based on usage patterns, rather than staying fixed forever
• Agent Assembly Pattern — Knowledge + Strategy + Execution + Verification in a complete AI Agent paradigm, not just a toolchain
• Ready Frontier Engine — Execute as soon as dependencies are met; no waiting for same-level peers
• TypeScript Native + Embeddable — Not a standalone service; npm install into any React app

npm install @opcpflow/core @opcpflow/nodes @opcpflow/react

---

Packages

@opcpflow/core       DAG types, validation, topology, Ready Frontier engine, ContextStore, EventBus, D4 evolution
@opcpflow/nodes      11 AI orchestration node definitions (colors, icons, categories, form fields)
@opcpflow/react      DAGEditor visual editor + sandbox execution
@opcpflow/engine     Ready Frontier execution engine, ContextStore, sub-graph replanning, telemetry

Dependency Flow

react → nodes → core
engine → core

---

Core Capabilities

Capability	Description
11 AI Nodes	trigger / task_decompose / dynamic / llm_call / api_call / mcp_tool / knowledge / strategy / verification / merge / output
Ready Frontier Engine	Dependency-activated execution; no level-based waiting
ContextStore 3-Level Memory	L1 scratch / L2 structured state / L3 external cache, with conflict detection + freshness + source tagging
EventBus	dag.* / node.* lifecycle events for observability and evolution hooks
D4 Evolution	Static → semi-static → dynamic → evolution reuse; unmatched sub-tasks automatically become sub-DAGs
D4EvolutionHook	Tracks dynamic node frequency and latency; auto-suggests promotion to static nodes
Token Budget Control	Track token consumption; circuit breaker on over-limit
Auto Data Routing	Connect edges = data flows; zero-config input/output
Sandbox Testing	Execute DAGs inside the editor; nodes change color in real-time
Headless CI	testDAG() for UI-less execution, ideal for CI pipelines

---

Node Types (11)

Category	Type	Description
control	trigger	DAG execution entry point
	task_decompose	Split commands into parallel sub-tasks
	dynamic	Catch-all handler + D4 evolution
	output	Final deliverable output
ai	llm_call	LLM inference / generation
	knowledge	Multi-source knowledge retrieval
	strategy	Persona / rules / behavior guidelines
integration	api_call	HTTP API calls
	mcp_tool	MCP protocol tool calls
	merge	Multi-type asset composition
verification	verification	SGV adversarial quality check

---

D4 Evolution Model

L1: Static               L2: Semi-Static          L3: Dynamic             L4: Evolution Reuse

Complete DAG             User draws skeleton +    No predefined DAG       Best-practice auto-
drawn by user            dynamic catch-all        fully dynamic           retrieval and adaptation
Fully predictable        Flexible                 Learning path           Continuous micro-evolution

           ───→ increased usage ───→ sub-task path stabilizes ───→ D4EvolutionHook suggests promotion

---

Quick Start

import { DAGEditor } from '@opcpflow/react'
import { createDefaultRegistry } from '@opcpflow/nodes'

const registry = createDefaultRegistry()

export default function App() {
  return (
    <DAGEditor
      registry={registry}
      onSave={(doc) => console.log('Saved:', doc)}
    />
  )
}

Programmatic Execution + D4 Evolution

import { DAGExecutionEngine, HandlerRegistry, D4EvolutionHook } from '@opcpflow/core'

const engine = new DAGExecutionEngine({ maxTokens: 100000 })
const evolution = new D4EvolutionHook()
const handlers = HandlerRegistry.createWithBuiltIns()

const report = await engine.execute(dag, handlers, {
  mode: 'live',
  onDAGComplete: (report) => {
    const insights = evolution.analyze('my-workflow', report)
    console.log('Suggestions:', insights.suggestions)
  },
})

console.log('Token usage:', engine.getStore().getTokenUsage())

---

Development

pnpm install
pnpm -r build
pnpm -r test
cd apps/demo && pnpm dev

License

Apache 2.0