Build production-ready, multitenant AI agents for your SaaS in minutes, not days, weeks, or months.
Anthropic and OpenAI give you the engine. Openloops gives you the chassis, with the AGI loop, the tools and everything an agent needs in production. You build the body: a customer support agent, a financial auditor, a meeting copilot.
New to AI agents? See what openloops is, explained with cars
Works with
OpenAI
Anthropic
Google Gemini
MongoDB
Sentry
Langfuse
Ollama01 / Live demo
Add an agent to your SaaS
02 / Features
What you get out of the box
Multitenant by design
Support multiple organizations, with users that belong to them. Chats, tasks, context and MCP servers are scoped from the start, so each customer gets their own agent space.
Zero-boilerplate persistence
Messages, tasks and every tool call are saved to MongoDB, and a conversation resumes exactly where it stopped, even in the middle of a confirmation.
Safe tool execution
Arguments are validated, destructive tools wait for the user's confirmation, and identifiers the model invents never reach a real system.
MCP with a trust flow
Connect any MCP server and its tools become regular tools. Each server is approved before its first tool runs and reviewed again when it changes.
Autonomous runs and automations
The same agent works without a person in the chat, on a schedule or when another system sends it an event, with a history of every run.
Guardrails
Input guards block card numbers, emails, code or anything you define before the model sees the message.
Usage and cost tracking
Every LLM call is recorded. Get tokens and cost per answer, per chat and per user.
Users and authentication
User management, login with refresh tokens and password reset are included, so you don't need a parallel system for identity.
Any LLM provider
OpenAI, Anthropic, Google, self-hosted models such as Ollama and private deployments on Bedrock, GCP or Azure work the same way, chosen per chat.
03 / Code
See how little code this actually takes
A marketplace of plugins for openloops
Extend your agents with ready-made packages instead of building every tool yourself. Install a plugin and your agent gets new abilities.
04 / Compare
How it compares to other frameworks
| openloops | LangGraph (OSS) | CrewAI (OSS) | |
|---|---|---|---|
| Persistence | Built in, zero config (MongoDB) | You choose and configure a checkpointer: in-memory, SQLite or Postgres, the last two in separate packages | Memory is opt-in and stored as local files (ChromaDB and SQLite, or LanceDB in newer versions), tied to the machine |
| Multi-user scoping | Built in: chats, tasks, context, and MCP servers are per-user from the start | Conversations are scoped by a thread_id. Mapping them to users, and scoping memory and tools per user, is up to you | Not a concept of the framework, you build it |
| Tool execution and retries | Validation, confirmation of destructive tools, duplicate and invented-id protection, and LLM and MCP retries are built into the harness | Per-node RetryPolicy and error handling that you configure | Retry limits are configured per agent or task |
| MCP support | Native: servers are a per-user record, tools load on demand, and every server goes through an approval flow | A separate package (langchain-mcp-adapters) with a client that you wire into the graph | Native mcps field on agents, or MCPServerAdapter. Servers are declared in code, not loaded per user |
05 / Modules
One package, organized by what you need
Agent, AgentLoop, RunContext, Harness: the runtime
AgiLoop, the default loop, and the skills loops share
Tool, BaseParams, and built-in tools such as web search and fetch_url
Skill, for LLM-backed reasoning steps inside a loop
Guard and ready-made input guards
MCP server management, connection checks and trust decisions
Automations: tasks the agent runs by schedule or by event, their runs and the runner that executes them
Chats and their messages: list, read, update, delete
User management
Login and token issuing
LLM usage and cost tracking
CurrentUser, CurrentSession
Logger and shared utilities
LLMClient and the HTTP clients used by built-in tools
06 / FAQ
Frequently asked questions
Is openloops free and open source?+−
Yes. Openloops is open source under the Apache 2.0 license and free for commercial use.
What do I need to run it?+−
Openloops is a TypeScript package. You install it with npm or yarn, and it saves messages, tasks and every tool call to MongoDB, so a conversation resumes exactly where it stopped.
Does it include an API or an HTTP layer?+−
No. Openloops is a package that your backend calls: it provides the use cases (running agents, chats, users, automations) and your backend exposes the endpoints and owns the authorization of each one. The integration guide lists which endpoint calls which use case.
Which LLM providers does it support?+−
OpenAI, Anthropic, Google and self-hosted models such as Ollama work the same way, and the model is chosen per chat. It also supports private provider deployments in your own cloud account, such as Anthropic on Amazon Bedrock, Google Cloud (GCP) and Azure.
Does it support MCP?+−
Yes. Connect any MCP server and its tools become regular tools. Each server is approved before its first tool runs and reviewed again when it changes.
How is openloops different from LangGraph or CrewAI?+−
Persistence, per-user scoping, tool safety, MCP and cost tracking come built in instead of being something you wire yourself. The comparison page lists the differences feature by feature.
Can I use it in a multitenant SaaS?+−
Yes. Openloops supports multiple organizations, and users belong to them. Chats, tasks, context and MCP servers are scoped from the start, and user management and login are included, so you can offer an agent to each of your customers.
Can I write my own loop?+−
Yes. Openloops ships AgiLoop, a general purpose loop that works out of the box, and you can write your own as an advanced option.
Ready to build your first agent?
Install openloops and have a working, persisted, multitenant agent running in minutes.
Not sure where to start? Read the explanation