Anonde Agent · Private beta

Free the data.
Use every model.
Secure every word.

Anonde Agent runs inside your network and anonymizes PII and secrets (names, emails, API keys) before it reaches Claude, ChatGPT, Cursor, or any model. Real values are revealed only inside your own trust boundary.

Join the waitlist → See how it works

In-network tokenization Any LLM backend Apache-2.0 engine

You don't change your LLM vendor. You don't add one to your audit scope. You add a PII boundary in front of what you already use.

In front of what you already use

Keep your model. Keep your contracts. Add the boundary.

Anonde sits between your people and whatever model you already trust. The upstream provider only ever sees tokenized text, so it never enters your audit scope as a new processor of personal data.

OpenRouter by default · BYO OpenAI, Anthropic, Azure OpenAI, AWS Bedrock · self-hosted vLLM & Ollama

How it works

Redact. Call the model. Reveal only inside your boundary.

01 · REDACT

Tokenize in-network

PII and secrets are detected and swapped for stable placeholder tokens before the prompt leaves your network.

02 · CALL

Forward only tokens

The tokenized payload goes to the model you chose. The provider sees structure and intent, never identities.

03 · REVEAL

Restore locally

The response is de-tokenized inside your network, gated by actor and purpose. Raw values never touch a third party.

The Anonde Agent

One gateway. Every AI tool covered.

The Agent is deployed inside your network and intercepts LLM-bound traffic at the egress, tokenizing PII before anything leaves. Desktop apps, IDE copilots, browser chats, CLI tools: every client that routes through it is covered, with nothing to install on employee machines and no behavior change. Sensitive data never crosses the boundary.

Get early access →

your network · Anonde gateway

Claude Desktopcovered

ChatGPTcovered

</> Cursor / Copilotcovered

$_ CLI toolscovered

↑ only tokens cross the trust boundary ↑

Live demo

Watch a real prompt get redacted.

Paste anything sensitive. PII is masked before the model sees it and revealed locally in the answer. The same engine that powers the Agent.

Send anything sensitive.

Started you off with a dispute case. Pick another sample or write your own. PII is masked before the model sees it and restored locally in the answer.

Why Anonde

Three things competitors can't copy in a week.

Best-in-class detection

A public, quarterly benchmark across languages and corpora. We publish wins and losses. Accuracy is a measured artifact, not a marketing claim.

View the benchmark →

Recognizer ecosystem

Community-authored recognizers per industry, region, and language. A searchable catalog that compounds with every contribution.

Audit-as-artifact

Cryptographically attestable, write-once logs that map to SOC 2, HIPAA, and EU AI Act Article 12: evidence your compliance team can hand an auditor.

Benchmarks

Lowest leak rate on 29 of 29 corpora.

Leak rate is the share of PII a data anonymization engine misses, so lower is better. In our latest public benchmark, Anonde leaks less than Microsoft Presidio across clinical, legal, finance, and PII corpora in five languages.

Anonde Presidio GLiNER PII

All corpora · Σ

Anonde10.1%

Presidio41.5%

GLiNER PII32.8%

Clinical / medical

Anonde11.2%

Presidio31.1%

GLiNER PII28.0%

General structured PII

Anonde12.5%

Presidio57.8%

GLiNER PII34.7%

Legal / administrative

Anonde2.5%

Presidio29.6%

GLiNER PII21.5%

Retail finance

Anonde4.9%

Presidio21.3%

GLiNER PII23.5%

Across the matrix, anonde-ner is the lowest-leak engine on 29 of 29 gold-annotated corpora. Rolled up, the Σ-all leak rate is 10.1% (Anonde) vs 41.5% (Presidio). The premium anonde-ner-stack image lands at 7.6%.

See the full matrix → · Anonde vs Presidio →

Built for GDPR HIPAA EU AI Act · Art. 12 SOC 2 Healthcare · Fintech · Legal

Private beta

Be first on the Anonde Agent.

Leave your name and email. We'll reach out when access opens for your team.

Name + email, or continue with GitHub.

Open source

The engine is free and self-hostable. Forever.

The Anonde detection-and-tokenization engine is Apache-2.0, written in Go, and runs as a library or a container on your own infrastructure. The Agent and the admin, SSO, and audit platform are the commercial layer on top.

Star on GitHub Read the docs

Talk to us

Building AI on regulated data?

Book 15 minutes with the founder. No pitch. We want to hear what's blocking your AI use case and what would make Anonde genuinely useful to you.

Book a call hello@anonde.io

Free the data.Use every model.Secure every word.