Skip to content
Comparisons

Is free-pi safe for your code? What freepi.ai says it does with your sessions

free-pi is free because ads and training data pay for it. Its own consent text says your code may be sold, with no opt-out.

In short

  • Everything below is quoted from free-pi’s own docs and privacy policy, checked on November 17, 2026.
  • free-pi saves your prompts, code, files and model responses, trains models on them, and says they "may be shared or sold to third parties".
  • Once you accept, there is no opt-out "except where the law gives you one", and no self-service way to delete your data.
  • An ad partner gets excerpts of your recent conversation, and free-pi keeps your full IP address indefinitely.
  • Fine for public or throwaway code. A bad fit for client work, employer code, or anything with secrets in it.
  • free-pi is three months old, calls itself "an early experiment", and has no independent coverage we could find.

free-pi (freepi.ai) is a free terminal coding agent built on the Pi harness. People ask whether it is safe, and the honest answer depends on what you mean. The CLI is open source and the company is unusually direct about its data terms. The terms themselves are the issue. Here is what they say.

Your sessions become training data, and may be sold

free-pi’s ads and privacy doc summarizes the consent screen you see on first run. Your coding sessions, which it defines as prompts, code, files and model responses, "are saved as traces and used to train models". It adds that they "may be shared or sold to third parties".

The privacy policy, from DappHero Corp, says the same in bold and notes that traces can include "proprietary code", credentials, API keys and passwords if they show up in your session. free-pi tries to strip personal information first, but says it "cannot promise" a trace is perfectly clean.

There is no opt-out and no delete button

After you accept, the docs say there is "no opt-out, except where the law gives you one". There is also "no self-service deletion yet". To ask about your data you email legal@freepi.ai with the subject "Privacy Request". If you decline the consent screen, the CLI exits, so using free-pi at all means agreeing to this.

Ads see your recent conversation

Some free-pi ads come from a third-party advertising partner. To pick an ad, free-pi sends that partner "a short excerpt of your recent conversation", meaning your latest prompts and the model’s replies with detectable secrets removed, plus your IP address, user agent and a hashed account ID.

Other things in the policy worth knowing

  • free-pi stores your full IP address with your account and keeps it "indefinitely".
  • Redacted excerpts of your recent prompts go to a third-party AI service that labels how you use the product.
  • The service calls itself an alpha, with "no warranty or uptime guarantee".
  • One CLI session per account at a time, and a daily allowance whose size isn’t published.

So is free-pi safe?

How much is riding on a three-month-old service

Trusting a tool with your code is partly a bet on who runs it. free-pi’s domain was registered in August 2026 and its CLI first reached npm on August 14, 2026. It calls itself "an early experiment", its allowance isn’t published and "can change", and we found no independent reviews or news coverage of it on November 18.

Freebuff has been live since February 2026. It is built by the team behind Codebuff and backed by Y Combinator. More than 850,000 developers have signed up, and it has been written up by RuntimeWire, Bitdoze and Carbon Ads. Its CLI was downloaded 106,189 times from npm in the 30 days to November 16, about five times free-pi-cli’s 20,720.

We found nothing suggesting free-pi is malware, and it tells you plainly what it does. If you’re learning, hacking on a side project, or working on code that is already public, the trade may suit you. If the code belongs to an employer or a client, you probably don’t have the right to let it be trained on and sold, and free-pi’s own docs tell you not to put such code in a session.

How Freebuff handles the same question

Freebuff is also free and ad-supported, so it isn’t data-free either. The difference is that you get a choice. Models marked "May use data for AI training" in the picker may keep your submissions for training, and Freebuff may share or sell that data. Models without the label aren’t used for AI training. We do use messages and usage to personalize ads and train the models that pick ads, and project files and connected repositories aren’t given to advertising providers. The details are in our privacy policy.

Freebuff also publishes its allowance: up to 30 hours a day in the US, in hour-long sessions with unlimited messages. For a full side-by-side, read Freebuff vs free-pi.

A free agent where you choose what trains

No card, no API key. `npm install -g freebuff`

Get Freebuff

Questions

Is free-pi safe to use?
It isn’t malware, and its CLI is open source. But its terms say your prompts, code and files are used to train models and may be sold to third parties, with no opt-out after you accept and no self-service deletion. Don’t use it on code you aren’t allowed to share.
Does free-pi sell my data?
Its docs say your coding sessions "may be shared or sold to third parties". Its privacy policy says the same, and adds that some ads are chosen by a partner that receives excerpts of your recent conversation.
Can I opt out of free-pi training on my code?
Not after you accept, except where the law gives you a right to. If you decline the consent screen, the CLI exits and nothing is sent.
How do I delete my free-pi data?
There is no self-service deletion. free-pi asks you to email legal@freepi.ai with the subject "Privacy Request".
Is there a free alternative that doesn’t train on my code?
Freebuff is free and lets you pick models that aren’t used for AI training; only models labeled "May use data for AI training" are. Freebuff does use your data to personalize ads and train ad models.
Is free-pi legit?
It is a real product from DappHero Corp, but a young one: launched in August 2026, described by its own homepage as "an early experiment", with an unpublished allowance and no independent coverage we could find.