@operad/core

Event-sourced graph runtime for AI agents. Every mutation is an event. Every event has a cause. Every decision is recorded.

Why

AI agents forget what they did, can't explain why, and don't self-correct. Operad fixes this with an event-sourced graph where every action produces an immutable event with causal chains, every decision records alternatives and confidence, and staleness detection keeps knowledge fresh.

Quick Start

npm install @operad/core @operad/adapter-memory

import { createRuntime, behavior } from '@operad/core'
import { MemoryAdapter } from '@operad/adapter-memory'

const runtime = createRuntime({
  storage: new MemoryAdapter(),
  behaviors: [
    behavior({
      name: 'log-claims',
      on: ['object.created'],
      handler: async (event, graph) => {
        console.log('New object:', event.payload)
      },
    }),
  ],
})

const graph = await runtime.createGraph('my-agent')
await graph.addObject({ type: 'claim', data: { amount: 5000 } })

Try the demo

npx @operad/server demo primitives

Core Primitives

| Primitive | What it does | |-----------|-------------| | Graph | Objects (nodes) + relations (edges) with typed data | | Event Log | Immutable, append-only with causal chains (causedBy) | | Behaviors | Reactive subscriptions — fire when events match | | Decisions | Recorded choices with alternatives and confidence scores | | Health | Staleness tracking — know when knowledge is stale | | Patterns | Cypher-subset queries — find structures in the graph | | Patches | Governance gates — propose changes, require human approval |

Storage Adapters

• @operad/adapter-memory — In-memory (dev/test)
• @operad/adapter-postgres — PostgreSQL (production)

Links

• GitHub
• Blog
• Getting Started

License

MIT