cortex-orchestrate

Memory-first multi-agent orchestration SDK — built on mem0.

npm install cortex-orchestrate

Quick start

import Cortex from 'cortex-orchestrate'

const cortex = new Cortex({
  apiUrl: 'https://your-cortex-api.onrender.com',
  userId: 'user-123',
})

const canvas = cortex.canvas('job-screener')

// Run an agent pipeline
const result = await canvas.run({
  trigger: 'Evaluate candidate: Jane Doe, 5 yrs Python, AWS certified',
})

console.log(result.output)
console.log(`Memories written: ${result.memoriesWritten}`)

// Fetch accumulated memories
const memories = await canvas.memories()

// Search memories semantically
const relevant = await canvas.search({ query: 'Python candidates', limit: 5 })

How it works

Every canvas in Cortex is a named memory namespace (userId::canvasId). Agents write to and read from this namespace across runs — so each execution is informed by everything that came before it.

Run 1 → cold start → agents evaluate → write memories
Run 2 → warm start → agents recall Run 1 patterns → write delta memories
Run N → deep context → quality compounds

API

new Cortex(config)

| Option | Type | Description | |--------|------|-------------| | apiUrl | string | Your Cortex API base URL | | userId | string | User identifier for memory scoping | | apiKey | string? | Optional bearer token |

cortex.canvas(id)

Returns a Canvas instance scoped to that id.

canvas.run(options)

Executes the agent pipeline with the given trigger string. Returns { executionId, output, memoriesWritten, agentsRun }.

canvas.memories()

Returns all memories accumulated in this canvas namespace.

canvas.search({ query, limit?, agentId? })

Semantic search over canvas memories.

canvas.clearMemories()

Deletes all memories in this canvas namespace.

License

MIT © 2026