Introducing AI Credits and Extra Usage

TLDR;

We are introducing three updates to help you spend more money get the most out of Fibery using AI, automations, and integrations:

1. AI is now measured in credits instead of calls: simple queries cost less, complex ones cost more.
2. In case you run out of AI credits, automation runs, or integration syncs included in your Pro plan, you can now purchase more without upgrading to Enterprise.
3. To help you decide whether you need extra credits or runs, we’re adding a usage dashboard.

Whether you are suspicious or curious, here’s the context behind those updates.

Problem for future us

We were ahead of the curve when we ditched the $10 user/month AI add-on and included AI into regular pricing plans back in November 2024. The rest of the industry (incl. Notion and Airtable) followed but not until mid-2025.

However, since we were so early, we had virtually no historical data to estimate the AI limits. Moreover, there was no industry standard of even what to charge for. So we chose the easiest metric to explain — AI calls aka answers — and pulled some numbers out of thin air (e.g., 100 calls per month on Pro plan). The AI adoption was tiny, so the limits didn’t matter much. “It’s a problem for future us to solve”, we said.

Well, the AI adoption has ramped up since, and so has our Anthropic bill. Unfortunately, we are now the future us 😅.

Terrible currency

The number of AI calls has turned out to be a terrible proxy metric for estimating both value delivered and costs incurred. Why? The calls are widely different.

Currently, both “Is there a dark theme in Fibery?” and “Suggest three hypotheses why churn rate increased in June based on the data in our CRM” consume the same single call despite the second interaction being a lot more valuable for the customer and costing 100x as much.

In fact, when plotting AI calls’ complexity, we have to resort to using logarithmic scales on both axes 😱 — such is the variability:

This level of variability was unseen in November 2024 when 4o was all the rage, and even the frontier models lacked reasoning and couldn’t use tools.

By contrast, now we see individual Fibery AI calls taking 30+ minutes and costing upwards of $10 🥲.

AI credits to the rescue

So we had to come up with a new meaningful metric and calculate monthly limits to include in each plan that would be both fair and sustainable.

The metric should be model- and even provider-agnostic as we juggle models under the hood depending on the task to achieve the best results. This means we can’t really use tokens as our currency because Haiku tokens ≠ Opus tokens ≠ GPT-5.5 tokens.

Luckily for us, the industry has converged on using AI credits as an abstraction on top of tokens/complexity/cost. All we had to do now is to calculate how many credits to include in each plan assuming that the simplest meaningful task should cost about 5 credits.

We crunched the numbers. They weren’t pretty…

Saving your money

It turned out, our AI agent was quite ineffective, so we would be able to offer considerably fewer credits than our competitors. No good.

So we made our AI considerably more cost-effective:

Why hadn’t we done this earlier? You see, it’s one thing to pay an exorbitant Anthropic bill every month, but forcing you to do so is unacceptable:

That’s our shame-driven development at its finest 🤌.

AI credits to the rescue (attempt #2)

After the optimizations went live, we arrived at the numbers we are comfortable with. Each month, depending on the plan and the number of seats, you’re getting a bunch of credits included with your subscription:

• Enterprise: 1000 AI credits per paid seat;
• Pro: 500 AI credits per paid seat;
• Standard: 250 AI credits per paid seat;
• Free: no credits 👿.

Note that our MCP doesn’t consume any Fibery AI credits, you can use it for free without limits. At least, for now 😈 .

How does this compare to your previous limit, you might ask? That depends heavily on the number of paid users in your workspace. Assuming a median call is worth ~18 credits, here’s the comparison:

Larger customers will get slightly more AI included, while smaller ones will, unfortunately, get less. That’s because we had to get rid of the base allowance: 5000 credits + 500 per paid seat.

You see, since the cost price of each credit is real, we have to hard-tie the caps to the number of paid seats. Otherwise, the economics would become unsustainable with some customers costing us more in AI bill than they bring in revenue. In fact, with our reckless old limits, such customers accounted for half of our AI bill 🙈.

Going beyond limits

That being said, if you’re living the future and have more AI agents than employees, who are we to stop you?

If you find yourself in the middle of a month no AI credits left, you can now enable extra usage. Set a monthly spending cap (e.g. €30 per month) and continue your normal operations.

Once a month is over, you’ll be charged only for the amount used (e.g., €20.48), not the entire cap. There will be an extra invoice in your billing settings.

Soon, you’ll be able to enable extra usage for automations and integrations as well. Here are the prices:

• €15 per extra 1000 AI credits;
• €5 per extra 1000 automation runs;
• €0.5 per extra 1000 integration syncs.

The minimum cap is €5 because transaction fees 🤷.

Note: if you’re billed in USD, just swap € for $.

Understanding usage

To help you decide how much extra usage you need (if any), we are introducing a new settings page with a bunch of analytics:

Check out how many credits, runs, and syncs you typically spend, and decide if you need some extra room.

Or notice that 94% of all your automation runs are consumed by a single crazy unoptimized Rule and fix it. It’s gonna feel great!

Fun future

The previous call-based AI billing has prevented us from offering complex things like vibe-coding custom apps on top of Fibery. It would’ve been too easy to spend more than €20 on 125 calls one-shotting interfaces for your app bankrupting our ass.

There was also no point in introducing a model switcher. Since both Haiku and Opus used to cost the same 1 call, why not always choose the best model available?

With AI credits, everything is up to you. Wanna spend extra €100 on credits to build a custom app that would replace a €500/month SaaS you hate? Go for it! Wanna switch to Haiku for a dumb question? Sure, it will barely cost any credits.

Discuss!

We hope that these updates will help you to extract the most out of Fibery while being more in control.

If you have any questions, curses, or compliments for us, join the discussion in our community.