openloops

Open source, Apache 2.0

Build production-grade enterprise agents in minutes, not days, weeks, or months.

openloops handles persistence, tool execution, MCP integration, and user management, so you can focus on the one thing that makes your agent yours: how it thinks.

Quick Start View on GitHub

npm install openloops

One line decides what your agent does

import { Agent } from 'openloops/core'
import { Logger } from 'openloops/common'
import { PlanExecuteLoop } from 'openloops/loops'
 
const logger = new Logger()
const agent = new Agent({ loop: new PlanExecuteLoop() })
 
const result = await agent.run({
  input: { message: 'what is the current price of Bitcoin?' },
  currentUser,
})
 
logger.info(result.data.lastAnswer, '[Demo] Result')

Launching this week

A marketplace of ready-to-use loops

Don't write a customer support agent from scratch. Pick a loop built for your use case and swap it in with one line.

ReactLoopCodeSpecialistLoopCustomerSupportLoopSecurityAnalystLoopFinanceLoop

What you get out of the box

Everything an agent needs to survive in production, already built

Multi-user by design

Chat history, task state, and MCP configuration are scoped per user from the start.

Zero-boilerplate persistence

Messages, tasks, and every tool call are saved automatically to MongoDB.

Production-grade execution

Retries, confirmation prompts, and crash recovery are built into the runtime.

Native MCP support

Connect any MCP server and its tools become regular tools instantly.

Sub-agent orchestration

An agent can launch and coordinate other agents to split up a problem.

Any LLM provider

OpenAI, Anthropic, and self-hosted local models all work the same way.

Works with

OpenAI

Anthropic

MongoDB

Sentry

Langfuse

Ollama

See how little code this actually takes

Running an agent, adding hooks, setting an identity, and writing a custom tool

import { Agent } from 'openloops/core'
import { Logger } from 'openloops/common'
import { CurrentUser } from 'openloops/base'
import { PlanExecuteLoop } from 'openloops/loops'
 
const logger = new Logger()
const currentUser = await CurrentUser.asyncFromDB(userId)
const agent = new Agent({ loop: new PlanExecuteLoop() })
 
const result = await agent.run({
  input: { message: 'hi, how are you?' },
  currentUser,
})
 
logger.info(result, '[Demo] Result')

One package, organized by what you need

openloops ships as a single package with subpath exports, so you only pull in what you use.

openloops/core

Agent, AgentLoop, RunContext: the runtime

openloops/loops

Built-in loops, starting with PlanExecuteLoop

openloops/tools

Tool, BaseParams, and built-in tools

openloops/skills

Skill, for LLM-backed reasoning steps inside a loop

openloops/mcps

MCP server CRUD and connection handling

openloops/users

User management

openloops/llmcalls

LLM usage and cost tracking

openloops/base

CurrentUser, CurrentSession

openloops/common

Logger and shared utilities

openloops/clients

HTTP clients used by built-in tools

openloops/authentication

Authentication utilities

openloops/chats

Chat and conversation persistence internals

How it compares to other frameworks

An honest comparison against the self-hosted, open-source versions of LangGraph and CrewAI

	openloops	LangGraph (OSS)	CrewAI (OSS)
Persistence	Built in, zero config (MongoDB)	Requires configuring a checkpointer (e.g. Postgres) yourself	Local memory (ChromaDB + SQLite) by default, machine-bound and scoped to a single run
Multi-user scoping	Built in: chats, context, and MCP servers are per-user from the start	Not a concept of the framework, you build it	Not natively supported by default
Tool execution and retries	Built into the runtime	You write the retry and error-handling logic	Configured per task, not automatic
MCP support	Native: MCP tools become Tool instances automatically	Requires a separate adapter package and manual wiring	Available through a separate tools integration, not automatic per-user loading
User management	Included (openloops/users)	Not included, you build your own	Not included
LLM usage and cost tracking	Included (openloops/llmcalls)	Not included in OSS, available via paid LangSmith	Not included as a first-class module
Observability	Sentry and Langfuse, enabled via env vars	Requires your own tracing setup, or paid LangSmith	Requires a third-party tracing integration
Pre-built agents	Growing marketplace of ready-to-use loops	You build every agent from primitives	No built-in agent marketplace
What you write	Node functions (plain async functions)	Nodes and the graph's edges, explicitly	Agents, tasks, and crew orchestration, declared per role
Mental model	A state machine your loop drives itself	A graph you declare and compile ahead of time	A crew of role-based agents coordinated for you

Ready to build your first agent?

Install openloops and have a working, persisted, multi-user agent running in minutes.

npm install openloops

Apache 2.0, free for commercial use

TypeScript, works with any LLM provider