One gateway between your app and any LLM. Detect, sanitize, rehydrate in a single call. Typically around 150 ms.
One gateway between your app and any LLM. Detect, sanitize, rehydrate — typically around 150ms, your pipeline never waits.
Send us text. We detect and replace sensitive data with safe placeholders. Typically around 150ms. Your pipeline doesn't wait.
Send the safe text to Claude, GPT-4, Gemini — whatever you want. The model sees placeholders only — never real data.
Send us the LLM response. We restore original values. End users see real names. The LLM never did.
Same detection pipeline. Two outputs — pick per request.
Replace sensitive data with safe placeholders. Send to any LLM. Restore the originals after. The LLM never sees real data.
Session-based PII mapping · never stored permanently
Sensitive data is permanently removed. No session. No recovery. Ship to auditors, store in logs, hand to third parties.
Nothing to leak · nothing to restore
Names, contact info, financial, healthcare, location, network identifiers — plus your own custom patterns.
Employee IDs, project codes, internal reference numbers — anything your organization needs.
Read the docsOne import, one call. Native SDKs and a drop-in proxy for OpenAI, Anthropic, and Gemini.
pip install ambientmeta
pip install langchain-ambientmeta
pip install llamaindex-ambientmeta
Audit every detection event. Keep data on your own infrastructure.
Audit every detection event by entity type, confidence score, and timestamp.
Session data is ephemeral; PII mapping is never stored permanently.
Deploy AmbientMeta on your own infrastructure. Single Docker image. Your data never leaves your network.
Run fully air-gapped. Same API, same SDKs — no data leaves your perimeter.
Need data sovereignty? Keep every byte inside your perimeter.
Start free. Pay as you grow. No credit card required.
For trying it out
For solo builders
For growing teams
For high volume
For regulated and at-scale teams: custom volumes, self-hosted or air-gapped deployment, and a dedicated support relationship.
All plans include all 17 entity types, sanitize + redact, and the full detection pipeline.
Yes. 1,000 requests/month, all entity types, full API access. No credit card required.
Yes. Single Docker image, self-hosted or air-gapped. Same API, same SDKs — no data leaves your network.
Typically around 150ms on the detection path. Fast enough for production LLM workflows without meaningful delay.
One call to /v1/sanitize or /v1/rehydrate = one request. Calls to /v1/patterns and /v1/feedback are free and unlimited.
Still have questions? Read the docs or talk to sales.
Create your account in 30 seconds. No credit card required. Read the docs first if you'd rather.
Ship with data protection this afternoon.