Which AI agents don't train on your data?

Casama, Hark Pro, Lindy and n8n say they don't use your data to train AI models, and Manus says so for team plans. Instinct, Meta Muse and Claude's consumer plans train by default until you switch it off. Self-hosted agents have no cloud of their own, so your model provider's policy applies. Several products don't say either way.

Updated

The list

AgentTraining on your data
CasamaNot used to train models
Hark ProSays it doesn't train on your data
LindyNo, says Lindy
n8nSays assistant data isn't used to train AI providers' models
ManusTeam plans say no; individual plans are unclear
Grok BotA training opt-out exists
InstinctOn by default; you can opt out
Meta MuseOn by default; you can turn it off
Claude Cowork and DuetOn for consumer plans unless you switch it off
OpenClawThere is no OpenClaw cloud; your model provider's policy applies
Hermes AgentWith your own keys only your provider sees your data; Nous Portal may train on prompts unless Privacy Mode is on
NanoClawNot stated; self-hosted data stays on your machine
Agent ZeroNot stated; it runs on your machine and sends prompts to the model you choose
nanobotNot stated
RakazoNot stated either way
MausBotNot stated
OpenBotNot stated for OpenBot

Checked on November 18, 2026, from each product's privacy pages and terms; the sources are at the end. We didn't find a statement for ChatGPT Dots. We make Casama.

Two layers to check

An agent usually has two companies behind it: the one that makes the agent, and the one that runs the AI model. Both see your requests.

  1. The agent maker. Do they use your chats, files and memories to train anything? Is that on by default? Is the answer the same on free and paid plans?
  2. The model provider. When the agent sends a step to a model, that provider's terms apply. With your own API key you choose the provider and can read its terms. With a hosted agent, ask which providers it uses.

A local model removes the second layer: nothing leaves your machine.

Training isn't the only question

A product can promise no training and still keep your data longer than you'd like. Also check:

Your data in Casama

See how Casama keeps your data safe.

Questions

What is the most private AI assistant?

One you run yourself on a local model: nothing leaves your machine. Among hosted ones, start with those that say they don't train on your data, then check who can access it and where it is stored.

Does a training opt-out cover what I already sent?

It depends on the product. Its terms should say whether switching training off covers what you already sent or only new conversations. If they don't say, ask before you rely on it.

Do open-source AI agents train on my data?

The agent software itself usually has no cloud to send data to. Your model provider's terms decide what happens to your prompts, unless you use a local model.

Sources

Join the waitlist Run it on your own server