The best tool for removing identifying details from Hebrew text.
A text can give someone away without naming them. Ploni finds that too.
Tested against 16 tools, it found up to 6× more identifying descriptions, with a false-alarm rate up to 3× lower. Against the tools with the same privacy: about half the damage to the text.
Before your text goes to AI, Ploni finds what identifies people: names and numbers, plus details like job, place and age that together are enough to tell who someone is. You keep a text you can still work with.
No name, yet clear who it is
The only doctor in a small kibbutz in the Arava, aged 58, asked us for help summarising the interview. No name, and everyone there knows who it is.
Ploni finds these details, and you choose what to hide.
- Hosted in Israel
- Text never stored
- A full page in a third of a second
- Also on your own servers, fully offline
Descriptions that identify someone, even with no name
This text has no name in it, yet anyone who knows the area knows who it is. Ploni finds that too.
An interviewee described by role, place and history, with no name given.
Personal details: names and numbers
Names, phones, emails, addresses and ID numbers.
A support bot answers a customer charged twice.
Three steps before every AI request
PII removal, redaction, de-identification, anonymization: it goes by many names. With Ploni it takes three steps, and the identifying details come out before the text reaches the model.
Clean
Send Ploni your text. Personal and identifying details become numbered placeholders like [name_1].
Ask
Send the clean text to any model you choose. The model gets the question without knowing who it is about.
Restore
Put the original details back into the model's answer. We don't store your text, and the replacement map stays with you.
POST /clean { "text": "..." }
→ { "text": "...", "map": { ... } }
POST /restore { "text": "...", "map": { ... } }
→ { "text": "..." }
What it looks like
Texts we wrote as samples, with invented details. The results are the model's output as it came out, at the strict setting: yellow is what it hides, [blue] is what passes on. Ploni works in Hebrew today, so the samples are in Hebrew.
A lawyer asks for a demand letter for a client dismissed after announcing her pregnancy.
A doctor asks for a visit summary; the medical facts the summary needs stay.
A call-centre transcript goes for a summary and a claim recommendation.
An accountant asks for a letter to the tax office; the amount stays.
What other tools miss, Ploni finds
Why no names?
Most leading providers restrict publishing comparisons with their tools, so we don't name them. Don't take our word for it: test it free on your own text, against any tool you like. Found a tool that works better? We train Ploni free until it wins, and give you $2,000 in credit as a gift. Full terms
The tools that protect the most get there by hiding more. Ploni understands which words point to a person, so it finds more, hides fewer words it didn't need to, and leaves what requests usually need. Finding more descriptions is what separates Ploni from the big cloud tools, which leave 9 to 32 of every 100 people identifiable. Measure by measure:
- BetterDescriptions that give a person away: Ploni finds 89 of every 100, up to 6× more. Commercial tools: 15 to 58.
- BetterFalse alarms, words hidden for no reason: 12 to 15 of every 100 things Ploni marks to hide, up to 3× fewer. Commercial tools: 21 to 39.
- Better than the leading privacy toolsTexts that can no longer be used: on Strict, on our general test, 3 of every 100, against 7 to 13 with the leading privacy tools.
- BetterTexts where the answer gets noticeably worse: on Strict, about half as many as with the leading privacy tools.
- Better than the big cloud tools, on par with the leading privacy toolsPeople a strong AI could still identify in cleaned text: far ahead of the big cloud tools (one 9 to 23, the other 32 of every 100). On par with the leading privacy tools on both test sets (general: 2 of 205, against 2 of 205 and 3 of 210; rare types: 0* of 178, against 0 of 181 and 0 of 190).
- Better than the big cloud tools, on par with the leading privacy toolsNames and numbers missed: 0.1 to 0.3 of every 100. Far ahead of the big cloud tools (one 7 to 13, the other 32 to 40). On par with the closest tool.
Ploni's figures are on Strict, our strictest setting. The leading providers are the commercial tools we tested; most leading providers restrict publishing comparisons, so they appear by number.
Ploni against 5 of the 16 tools we tested
| Tools | People a strong AI could still identify (measured results) | Texts ruined or damaged by the cleaning | Names and numbers found | Descriptions that give a person away: found | False alarms: share of what it marks to hide |
|---|---|---|---|---|---|
| Ploni | General: 2 of 205 Rare types: 0* of 178 AI requests: 0* of 180 | 20–24% | 99.7–99.9% | 89% | 12–15% |
| Leading provider 1 | 9–23% | 19–36% | 87–93% | 27–28% | 22–31% |
| Leading provider 2 | General: 2 of 205 Rare types: 0 of 181 | 40–55% | 98–99.9% | 54–58% | 21–27% |
| Leading provider 3 | 32%† | 28%† | 60–68% | 15–16% | 38–39% |
| Leading provider 4 | General: 3 of 210 Rare types: 0 of 190 | 39–49% | 99.8–99.9% | 45–52% | 31–36% |
| HebSafeHarbor (open source) | 8%‡ | 28%‡ | 93% | 24–31% | 13–16% |
People a strong AI could still identify (measured results)
- PloniGeneral: 2 of 205
Rare types: 0* of 178
AI requests: 0* of 180 - Leading provider 19–23%
- Leading provider 2General: 2 of 205
Rare types: 0 of 181 - Leading provider 332%†
- Leading provider 4General: 3 of 210
Rare types: 0 of 190 - HebSafeHarbor (open source)8%‡
Texts ruined or damaged by the cleaning
- Ploni20–24%
- Leading provider 119–36%
- Leading provider 240–55%
- Leading provider 328%†
- Leading provider 439–49%
- HebSafeHarbor (open source)28%‡
Names and numbers found
- Ploni99.7–99.9%
- Leading provider 187–93%
- Leading provider 298–99.9%
- Leading provider 360–68%
- Leading provider 499.8–99.9%
- HebSafeHarbor (open source)93%
Descriptions that give a person away: found
- Ploni89%
- Leading provider 127–28%
- Leading provider 254–58%
- Leading provider 315–16%
- Leading provider 445–52%
- HebSafeHarbor (open source)24–31%
False alarms: share of what it marks to hide
- Ploni12–15%
- Leading provider 122–31%
- Leading provider 221–27%
- Leading provider 338–39%
- Leading provider 431–36%
- HebSafeHarbor (open source)13–16%
These are all 4 commercial tools and the open-source tool that found the most names and numbers. The other 11, all open source, found 4–67% of names and numbers and 0–32% of identifying descriptions. See the full table
Any tool can reach zero by hiding everything. The real question is how much privacy you get for how much damage.
You choose the setting. Ploni starts on Strict, and switching to Balanced takes one field in your request.
You also choose what to hide. If you don't choose, Ploni also hides places, workplaces, jobs, ages and exact dates. Medical and financial details, habits, events and family situation, among others, stay visible, because requests usually need them.
On our internal Hebrew benchmark against 16 tools, 4 commercial and 12 open source. "Leading privacy tools": the two commercial tools that leave the fewest people identifiable. "Big cloud tools": the two large commercial cloud services. †General test only. ‡Measured on one 100-document test sample. Balanced and the other tools as measured on 25 October; texts that can no longer be used are from the general test. See how we measured
Your text stays in Israel
Hosted in Israel
Text sent through the API is processed only in Israel, in the Google Cloud Tel Aviv region.
Never stored
We don't store or log the texts you send.
Fast
Short texts come back in under a tenth of a second, a full page in a third of a second. The first request after an idle spell takes several seconds.
On your servers
Available on your own servers, including fully offline, for organisations with strict requirements.
Can be trained on your texts
Further training on data you give us under a separate agreement, so Ploni learns the identifiers and descriptions of your field.
Hebrew now. More languages coming.
Amendment 13 to Israel's Protection of Privacy Law strengthened enforcement on personal data. What that means when you send client data to AI
The best Hebrew anonymization, guaranteed.
Test it free on your own text. Found a tool that works better? We train Ploni free until it wins, and give you $2,000 in credit as a gift.
"Better" means fewer identifying details left visible at similar text damage, measured with our comparison tool, on at least 50 texts or 10,000 words. Training runs for up to 12 months. The credit is used at normal rate limits. Full terms
Simple pricing, by the word
Pay as you go
$0.20 per 1,000 words
API hosted in Israel. We're in free early access, so nobody is charged now.
Larger organisations
Talk to us
Available on your own servers, including fully offline. Training on your data as an added project. hello@ploni.ai
Try it on your own text
Sign up with Google or email, get an API key and test right away. Business users only.