Freebuff Reviews

Summary: Overall experience: Freebuff used to be good, but now that they have more users, they have betrayed the free users and are moving towards subscriptions.

Positive: They USED to have models with unlimited usage but not anymore; now, everything will consume your quota.

Negative: ⚠️ WARNING: What Used to Be Free Now Effectively Requires a $60/Month Subscription
TL;DR
My experience with Freebuff has been extremely disappointing, and I would strongly recommend researching the service carefully before installing it and handing it access to your code or computer.

Here are the biggest problems:

Freebuff dramatically cut back its free usage. To roughly match what users previously received for free, you now need the $60/month subscription—and even that may not fully match the old allowance because additional limits now exist.

The actual coding quality and overall experience are far worse than the marketing makes them sound.

Their Terms require you to actively initiate and supervise Freebuff sessions and remain present while the agent works, which undermines the purpose of an autonomous coding agent in the first place.

Their privacy practices are extremely concerning, including AI-training provisions, advertising profiling, repository analysis, and third-party data sharing.

There have also been multiple incidents involving misleading marketing or platform-policy issues, including X, Product Hunt, Trustpilot, and OpenAI.

1. The Free Usage Was Cut Dramatically
Before
Free: approximately 180 premium sessions/month

Now
$25/month subscription: approximately 140 premium sessions/month

In other words, the $25 paid tier gives you fewer premium sessions than Freebuff previously gave away for free.

To get approximately the old free allowance, you now need the $60/month subscription.

And even then, you may still come up short because Freebuff now has additional restrictions such as token-spend limits.

Also, even cheap models like GPT 5.6 Luna are now gated behind at least the $25/month subscription.

So something users once received for $0 now effectively requires spending $720/year to approximately reproduce.

They Still Market Themselves Around “No Subscription”
One particularly frustrating part is that Freebuff has continued making claims along the lines of “no subscription”, despite there clearly being a paid subscription system.

The same applies to claims about “no account required” even though, in practice, an account is required.

That isn't the only example of marketing that I believe deserves scrutiny.

2. Their Own GitHub Documents Deliberately Reducing the Free Tier
This one is especially revealing because they literally admitted it (probably accidentally)—it appears directly in Freebuff's own source code.

Freebuff documented that it reduced the base free premium allowance when it introduced its “Levels” system.

Source:

https://github.com/CodebuffAI/freebuff/blob/main/common/src/constants/freebuff-models.ts

Look immediately above:

export const FREEBUFF_PREMIUM_SESSION_LIMIT = 4

The relevant comment explains that the base premium allowance was lowered when Levels were introduced and says:

“A free tier whose floor already hands out everything has nothing left to reward with...”

In other words, their own code comments acknowledge reducing the baseline allowance specifically so that additional usage could be used as a reward mechanic.

This is particularly ironic considering Freebuff has portrayed itself as being “deeply allergic” to these kinds of dark patterns.

3. The “Free” Product Looks Much Less Generous Once VC Economics Enter the Picture
Freebuff has VC/investor backing.

Likely case: The old generous free tier was useful for acquiring users, but once users were there, the economics changed substantially.

What was once approximately 180 premium sessions/month for free has become something that can require a $60/month subscription to roughly reproduce.

Whatever you call that strategy, users should know how dramatically the value proposition changed.

4. Their Terms Require Active Human Supervision
This is one of the strangest limitations for something marketed as a coding agent.

Freebuff's Terms of Service require users to:

initiate the session themselves;

supervise the session;

remain actively present while the agent is working.

That essentially defeats the purpose of an autonomous coding agent.

If I have to remain actively present supervising the agent the entire time anyway, what's the point?

Their Terms also give Freebuff broad authority to terminate accounts.

Given that I'm speaking up against them, they might do that to my account.

5. Coding Quality and Harness: Much Worse Than I Expected
Freebuff's model list can look impressive on paper.

The actual experience is another story.

Their custom coding harness is weak
Freebuff's custom harness/environment is, in my experience, significantly behind mature coding-agent environments.

It lacks several features that have become standard or extremely useful elsewhere, including:

Plan mode

Read-only modes

Plugin support

This matters more than people may realize.

The model itself is only one part of an AI coding agent. The agent harness—tools, context management, planning system, command execution, permissions, editing workflow, and feedback loop—can have an enormous effect on actual coding performance.

A great model in a weak harness can perform dramatically worse than expected.

Analogy: A great mind in a bad environment can perform worse than a lesser mind in a top-tier environment (like OpenCode).

6. Your Model Quality Depends on Where You Live
If you aren't located in one of Freebuff's preferred wealthy markets (US, UK, etc.), you will not receive the same models users elsewhere get.

You may instead be routed to bad models such as MiMo V2.5.

7. Frequent Bugs, Crashes, and Service Interruptions
I've encountered:

crashes;

bugs;

service interruptions;

unexplained failures;

sessions stopping unnecessarily.

The product simply hasn't felt reliable enough for something that is supposed to be trusted with meaningful software-development work.

8. It Frequently Stops and Makes You Continue Manually
Freebuff sometimes stops and asks you to continue even when there is literally no reason to stop.

Why I think they do this: forcing users back into the application increases attention to the ads (it's all profit driven)

9. Extremely Long Wait Times
Performance can also be absurdly slow.

At the time I wrote this review, I had an active Freebuff session showing a wait of:

>6,767 seconds (haha but I'm not joking)

That's roughly:

TWO HOURS

Just to get the agent started.

I'm not exaggerating.

Nearly two hours of waiting before the agent even begins working is not remotely acceptable for a coding tool I would depend on.

10. “DeepSeek V4 Pro” Doesn't Necessarily Mean the Current Version
Freebuff advertises access to DeepSeek V4 Pro.

But users should pay attention to the exact model version.

The version offered was an older DeepSeek V4 Pro release rather than the newer/current revision.

The difference between the two is day and night.
Simply displaying the broad model-family name can therefore make the offering look much more impressive than it actually is.

11. Models Can Be Quantized to Lower Quality
Some models Freebuff provides are quantized.

For people unfamiliar with that terminology: Basically it makes the model way worse but makes it cheaper (again freebuff is profit-driven and greedy)

So you may not necessarily be receiving the same quality you would get from the model's original/full-precision deployment.

In plain English:

You may be getting a cheaper, watered-down version of the model you think you're using.

12. Extremely Broad Command Execution + No Read-Only Mode
This is probably the issue I would take most seriously before installing Freebuff.

The agent is capable of running commands on your machine with basically unrestricted access.

If the model:

makes a serious mistake;

executes the wrong command;

is manipulated through prompt injection;

consumes malicious repository content;

or is otherwise compromised,

the consequences could be severe.

An agent with broad command execution could potentially:

delete files;

overwrite important data;

install malicious software;

modify system configuration;

leak secrets;

damage repositories;

execute malicious code;

absolutely cook your computer

And Freebuff does not provide a proper read-only mode to eliminate that risk when you simply want the agent to inspect a project.

That means even tasks that should require observation only can involve an agent with write/command privileges.

For a tool with this level of system access, I think the absence of strong permission controls is a major safety concern.

13. Privacy Practices That Users Should Read Very Carefully
This is another area where I strongly recommend reading the policies rather than trusting the marketing.

Freebuff can use prompt/message data for model training
Freebuff's policies allow certain personal prompt/message content to be used for AI/model-training purposes.

That alone would make me extremely cautious about giving the tool:

proprietary code;

private conversations;

credentials;

confidential business information;

sensitive repositories.

Solar Pro 4 AI-training disclosure
For Solar Pro 4, there was an AI-training disclosure issue where the relevant warning was supressed in the UI.

In other words, users could interact with the model without the AI-training warning being clearly presented to them.

That is exactly the kind of disclosure I believe should be impossible to miss.

14. Prompt and Message Content Can Be Processed for Advertising
Freebuff can send prompt/message content to its own servers and advertising-related providers for analysis and advertising personalization.

That means your interactions with a coding agent aren't necessarily isolated purely to generating code.

They can also feed into an advertising ecosystem.

For a developer tool that may routinely receive source code, technical descriptions, company information, and debugging data, that deserves serious consideration.

15. Data Can Be Retained “As Long as Needed”
Freebuff's policies allow information to be retained for as long as considered necessary.

Think carefully about what that means...

16. Connecting GitHub Gives Freebuff Crazy Repository Analysis Access
Connecting a GitHub repository authorizes analysis of repository contents.

That can include:

automated analysis;

human review;

Freebuff itself;

third-party service providers.

17. Paid Fake Social Engagement / Platform-Manipulation Marketplace (Got in trouble with X for this)
Freebuff promoted a marketplace/system involving paid artificial social engagement.

I reported the activity to X.

X subsequently determined that the reported activity violated its policies.

Whatever terminology you prefer—paid engagement, artificial amplification, or platform manipulation—I think users should know that a company selling an AI coding product was involved in promoting this kind of growth mechanism.

18. Advertising Profiling
Freebuff also engages in advertising profiling/personalization.

Again, that is an unusual thing to have to think about when choosing what appears at first glance to simply be an AI coding agent.

Users should ask themselves whether they are comfortable having coding-agent activity participate in an advertising-profile ecosystem.

19. Product Hunt Upvote Incentives
Freebuff rewarded users with additional AI usage for upvoting Freebuff on Product Hunt.

That creates an obvious conflict:

Users weren't simply independently deciding that Freebuff deserved an upvote—they had a direct economic/product incentive to give one.

That is, in my view, unethical manipulation of a supposedly organic ranking system.

It also appears inconsistent with Product Hunt's policies against incentivized voting.

So I would be very cautious about treating Freebuff's Product Hunt performance as purely organic evidence of user enthusiasm.

20. Trustpilot Currently Has a Warning on Freebuff For It
Trustpilot currently displays a warning associated with Freebuff indicating that its rating is unavailable because the company:

“is displaying Trustpilot content in a misleading way.”

Despite that, Freebuff has continued using Trustpilot reviews/content in its own advertising and marketing.

That's a pretty extraordinary warning for a company to have attached to its review presence, and I think prospective users deserve to know about it.

21. OpenAI Reportedly Blocked Freebuff's GPT-5.6 Luna Supply for a “Policy Violation”
Freebuff previously obtained GPT-5.6 Luna capacity associated with OpenAI.

That supply was reportedly blocked because of a:

“Policy Violation.”

Rather than simply dropping the offering, Freebuff subsequently obtained access through alternative providers.

Decide for yourself how you feel about that.

Personally, when an upstream model provider cuts off supply over a policy issue and the service then finds alternative routes to continue providing that same model, I think it deserves scrutiny.

22. Extremely Aggressive and Misleading Marketing
Taken individually, some of these issues might be explainable.

Taken together, I see a pattern:

“No subscription,” despite paid subscriptions existing.

“No account required,” despite an account being required in practice.

A formerly huge free allowance being dramatically reduced.

The old free level now effectively requiring a $60/month plan.

Free-tier reductions deliberately tied to engagement/reward mechanics.

Incentivized Product Hunt upvotes.

Trustpilot flagging the company's use of its review content as misleading.

Model-family names that may obscure older versions.

Quantized models being presented under familiar model names.

Geographic differences in model quality.

Privacy and advertising practices that are much broader than many users would expect from a coding agent.

A platform-policy violation involving paid/artificial social engagement.

An upstream AI-provider policy issue involving model access.

At some point, I stopped viewing these as isolated annoyances.

My conclusion
Freebuff initially looks incredibly attractive: premium coding models, lots of usage, and an apparently generous free offering.

Look deeper.

The free offering has deteriorated dramatically. The strongest historical allowance now effectively costs as much as $60/month, while additional limits may still apply.

The coding harness is missing important features. Reliability has been poor. Wait times can be ridiculous. Model quality isn't necessarily what the headline model names imply. Permissions are dangerously broad. Privacy and advertising practices deserve very careful scrutiny.

And several aspects of the company's marketing and growth tactics have raised serious concerns for me.

I would NOT recommend installing Freebuff or giving it access to a repository or machine you value without first reading its Terms, Privacy Policy, model disclosures, and permission model very carefully and deciding whether you are okay with the risks.

There are plenty of mature coding-agent alternatives.

Personally, after everything above, I would choose one of them instead.

Thanks for reading this review. I know it's a long one but there were so many things I wanted to say. I hope you now know not to use Freebuff. So here are the alternatives that I personally use:
- OpenCode: Currently has generous free Muse Spark 1.3 at time of writing (similar level as GPT 5.6 Sol)
- Cline: Currently has generous free GLM 5.3 Flash at time of writing (previously Ox Alpha)