morafree · no sales call · reply within 24hYour model needs data.
We'll find it.
Fine-tuning, post-training, a small model for one job. Tell us what it should learn from. We check what exists and reply within 24h. Free, and no sales call.
- 29B+rows indexed
- 3.7kdatasets checked
- 24hto an answer
- 0sales calls
- FineFineWeb1B+ rowstext-classification
- wikitext1M+ rowstext-generation
- fine-news100M+ rowstext-generation
- preprocessed_commoncatalog-cc-by10M+ rowsHugging Face
- mmlu100K+ rowsquestion-answering
- glue1M+ rowstext-classification
- yodas31M+ rowsaudio-to-audio
- GiftEvalPretrain1M+ rowstime-series-forecasting
- fusion-dw1M+ rowsHugging Face
- fineweb-edu-translated1B+ rowstranslation
- jat-dataset-tokenized10M+ rowsHugging Face
- super_glue100K+ rowstext-classification
- jat-dataset100M+ rowsreinforcement-learning
- evaluation-results100M+ rowsother
- us-stock-data100K+ rowsHugging Face
- common_voice_17_010M+ rowsHugging Face
- imdb100K+ rowstext-classification
What are you training?
The models teams bring to us, and the data each one learns from. Pick one to start from it.
- Conversations
Support agent
Support threads with the answer that solved them, tagged by issue
- Code
Code model
Pull requests with their review and the fix that followed
- Audio · Conversations
Voice agent
Real phone calls with transcripts, by language and accent
- Robotics · Video
Robot policy
Teleoperation runs: cameras, joint states and actions
- Agent traces
Computer-use agent
Screens, clicks and keystrokes of people doing real work
- Text · Tables
Domain model
Expert documents from one field: contracts, filings, notes
Where we look
Three places, in this order. You hear about all three in one answer.
- 01 · public
The public sets, already checked
3,747 datasets from Hugging Face and Kaggle, read for license, size and task. We tell you which allow commercial training and which do not.
Hugging Face
Kaggle - 02 · on Mora
Private data, licensed
Datasets companies list on Mora. Each one is checked: the company may license it, no personal data is left, and it is not already public.
licensedno personal datanot public - 03 · who holds it
The companies that have it
Most training data was never published. It sits in a startup's tools. We know which kinds of companies hold which records, and we ask them without saying who is asking.
Not a sales call. An answer.
- NowYou describe the model and the data. Two minutes.
- Same dayA person on the team reads it. Not a bot, not a sequence.
- Within 24hWhat exists, how close it is, its license, what is missing, a sample when there is one.
- ThenYou decide. No sales call, no follow-up sequence, no price to discuss yet.
{
"found": [
{ "where": "on Mora", "match": "close",
"what": "support threads with the fix, EN",
"license": "commercial", "sample": true },
{ "where": "public", "match": "partial",
"what": "open support dataset, no issue tags",
"license": "cc-by-4.0" }
],
"who_holds_it": "companies with this kind of record",
"missing": "issue-type labels",
"price": null
}What devs ask first
Will someone try to sell me something?
No. No sales call, no demo, no follow-up sequence. You get a written answer within 24 hours from someone who looked, and that is it unless you write back.
Is it really free?
Yes. There is no price yet, on purpose: we are learning what teams training their own models actually need. If something fits, we talk about the rest then.
What kind of data can you find?
Anything a model learns from: text, code, conversations, audio, images, video, robot runs, agent traces. Describe it the way you would to a teammate; we ask if something is unclear.
Is what you point me to licensed?
Every source comes with its license stated. Public sets carry the license their catalogue gives, and we tell you when it does not allow commercial training. Private data on Mora is licensed by the company that holds it.
What about personal data?
Private data on Mora is checked for personal data before anyone sees it: people are replaced by generated ids, the same id everywhere, so a conversation still makes sense.
Can I look before anything else?
Yes, when there is a sample. On Mora, the sample is drawn at random from the full dataset, so nobody can show you only their best records.
What if nobody has it?
We say so, plainly. Then we tell you who is likely to hold that kind of record and whether we can ask them for you.
Who sees what I describe?
Mora's team, nobody else. If we ask a company whether it holds the data, we never say who is asking.
Who is behind Mora?
The team behind PilotGPT, a speech model trained on air traffic control radio. We had to find and license our own training data, so we know how long that takes.
What does your
model need?
Two minutes to describe it. A written answer within 24 hours. Free, and no sales call.