Question 1

What happens to the data I enter into Flocta?

Accepted Answer

It is used to answer you, and for nothing else. We do not sell your data, we do not pass it to advertisers or data brokers, and we never train any model on it — not ours, not a partner's. Your account, your chats and your files are hosted in France, inside the EU, and the open-weight models we serve run on French infrastructure too. You stay the controller: we process strictly on your instructions under a GDPR data processing agreement, you can export or delete anything at any time, and the privacy notice names every provider and storage location involved. If you bring your own frontier plan, that provider's terms apply to those requests and we show you which request went where.

Question 2

How long does Flocta store my prompts and answers?

Accepted Answer

Exactly as long as you want. Your conversations live in your account, in France, so you can pick them back up, and you can delete any of them — or your whole history — at any moment. Delete a chat and it is gone for good: no shadow copy, no backup that quietly survives it, no archive we keep for training.

Question 3

Does Flocta train AI models on my data?

Accepted Answer

No. Never, on nothing. Not on your prompts, not on your answers, not on your files, and we do not hand them to a third party to train on either. That is a contractual commitment in our processing agreement, not a setting buried in a preferences page. The open-weight models we serve are already trained — running them on your input teaches them nothing.

Question 4

Which AI models does Flocta support?

Accepted Answer

All of them — that is the point. The frontier, from GPT-6 Astra and Fable 5.1 to the other leading proprietary assistants, sits next to the whole open-weight field: Kimi, GLM, Qwen, Mistral, Llama, gpt-oss and the rest. Frontier models come in through the plan you already pay for and move onto Flocta; the open-weight roster we serve ourselves in France, currently GLM 5.2, Qwen3.5 397B-A17B, Qwen3.6 35B-A3B, Qwen3 235B Instruct, Mistral Medium 3.5, gpt-oss 120b, Llama 3.3 70B Instruct, Gemma 4 26B-A4B, plus the enabled image models. When a new model ships, we build it in straight away, so you are never waiting on a release cycle. Concrete models stay directly selectable; Auto stays disabled until its routing targets have completed attestation.

Question 5

How much does Flocta cost?

Accepted Answer

We take a commission on the money our systems save you, and nothing else. No seat price, no subscription, no minimum. You bring the frontier plan you already pay for onto the Flocta platform, we move everything that does not need a frontier model onto cheaper models, and you pay us a share only of the spend we actually removed — after it is saved. No savings, no invoice. If you have no frontier plan at all, we offer a mixed plan with every model included; its price moves with your usage and stays below what any other LLM provider charges for the same work.

Question 6

We already run Claude company-wide. Can we use Flocta without switching?

Accepted Answer

Yes, and nobody has to migrate. Flocta is not a competitor to Claude — it is the layer underneath it. You transfer your existing Claude plan onto the Flocta platform and your team keeps working with Claude exactly as before. Behind it, Flocta sends routine work to cheaper open models and pulls in additional models for planning, execution and review when a task is genuinely hard. Same assistant, same habits, a much smaller bill — and you only pay us a share of what we save.

Question 7

We pay for ChatGPT across the company. Does Flocta replace it?

Accepted Answer

No — you bring it with you. Move your existing ChatGPT plan onto the Flocta platform and everyone carries on with the assistant they already know. Flocta decides underneath which model each task deserves: routine drafting, extraction and formatting go to cheap open models, hard work still goes to the frontier. Nothing to relearn, no migration project, and our commission comes only out of the spend we take off your bill.

Question 8

We already use Langdock. Why add Flocta?

Accepted Answer

Because Langdock is a place to work and Flocta is what makes that work cheap. Your Langdock plan moves onto the Flocta platform, the interface and workflows your team is used to stay where they are, and Flocta routes each request to the model that can actually handle it — open-weight where that is enough, frontier where it is not. You do not choose between the two; you run one on top of the other and pay us only from the savings.

Question 9

Our engineers live in Cursor. Does Flocta fit in?

Accepted Answer

Yes. Bring your Cursor plan onto the Flocta platform and your engineers keep their editor, their agents and their habits. Flocta shifts the bulk of the tokens — boilerplate, refactors, test scaffolding, log reading — onto cheaper models, and keeps the frontier for architecture, tricky debugging and review, where it earns its price. Coding is the workload with the most routine volume in it, so it is usually where the saving is largest.

Question 10

What is the best way to use Flocta?

Accepted Answer

Bring your frontier plan onto Flocta and keep working as you do today — that is the whole setup. From there two things open up: the portal for chat, files, company knowledge, images and the enabled voice features, and Flocta's routing underneath it, which decides per task whether a cheap open model, a strong open model or your frontier model should answer. If you would rather keep your assistant in the lead, you can also connect Flocta to it over MCP. Availability depends on the features enabled for your workspace.

Question 11

How good is Flocta's price–performance really?

Accepted Answer

Flocta keeps difficult planning and judgment with the frontier model and routes bounded execution to lower-cost open-weight models. Our current target is roughly 95% of the output quality on the workloads we delegate — and because we are paid only out of the spend we remove, we carry the downside when the routing is a bad fit. Real savings depend on your workload, your provider's limits and how much of your work is genuinely routine, so we measure it on your own tasks before anything is signed.

Question 12

Can Flocta replace my ChatGPT or Claude subscription?

Accepted Answer

It does not have to, and usually should not. The point is to keep the assistant your team likes and take the cost out from underneath it: move the plan onto Flocta, let routine work run on cheaper models, and drop to a lower tier only once your own numbers say you can. Validate that on your own tasks first — provider limits and routing suitability vary.

Question 13

Where is Flocta built and hosted?

Accepted Answer

Flocta is built in Berlin and hosted in France. The open-weight models we serve run on Scaleway's Paris infrastructure, so that processing stays in France. Your account, your conversations and your files stay on EU infrastructure under a GDPR data processing agreement, with deletion on demand, full provider disclosure and a hard no-training rule. Those are technical and contractual measures; the final legal assessment still depends on your organisation's purpose, contracts, configuration and compliance review.

Question 14

Which open-weight models are worth using in 2026?

Accepted Answer

The open field has closed most of the gap: Kimi, GLM, Qwen, Mistral, Llama and gpt-oss now handle the majority of day-to-day work at a fraction of frontier cost, which is exactly why routing pays. Auto is temporarily disabled pending model attestation. The concrete choices in this deployment are GLM 5.2, Qwen3.5 397B-A17B, Qwen3.6 35B-A3B, Qwen3 235B Instruct, Mistral Medium 3.5, gpt-oss 120b, Llama 3.3 70B Instruct, Gemma 4 26B-A4B, and new releases are added as they ship. Enabled image models handle image generation.

Question 15

Do I need AI expertise or API keys to get started?

Accepted Answer

No. There is no key to manage, no model to deploy and no infrastructure to run. Connect the plan you already pay for, open the portal and pick a model from the list. Auto becomes available once its targets pass attestation.

Question 16

Does Flocta support MCP or custom actions?

Accepted Answer

Yes. You can connect MCP servers and custom tools to give the models real actions — search the web, read a repo, hit your own APIs — so a chat can actually do things instead of only talking about them.

flocta +
The first cost-efficient enterprise
mixed-LLM platform in Europe.

Flocta — the cost-efficiency software layer for your product backend.

ControlEveryToken

Model agnostic

Company knowledge

Personalized voice inputs

Work with files

Company-Tailored Open-Weight Models

Private Cloud Model Deployment

More features

Future optimisation tracking

A tailored agent per task

Weekly voice reviews

Runs on your frontier plan

Token analytics and run reviews

Live monitoring

TrustedGermanStartup

Made in Europe

Backed by Atlantic VC

TokenCuttingIceberg

Too easy to fuck up.

Send the diff, not the history.

Fast enough not to retry.

Your own models, your cloud.

SystemsMadeVisible

TETFU Plan

ZERO WASTE CONTEXT

Don’t Think Twice

EverySingleDepartment

WhatFloctaSaves

How many team members have a frontier LLM plan?

QuestionsAndAnswers

Flocta — the cost-efficiency layer for the AI you already use, and Europe's first cost-efficient enterprise mixed-LLM platform. Built in Berlin, Germany.

flocta + The first cost-efficient enterprise mixed-LLM platform in Europe.

Flocta — the cost-efficiency software layer for your product backend.Flocta++ The cost-efficiency software layerfor your product backend.