Privedge
Open Source · MIT License

Deploy Privedge in your
own infrastructure.

The worker and the SDK are MIT licensed — fork the repo, run wrangler deploy in your own Cloudflare account, and own every node. No lock-in, no monthly cost, no data leaving your infrastructure.

View on GitHub Or use managed cloud →
Which option fits you

Self-host or managed cloud?
Pick what fits your team.

Self-Host MIT · Free forever
  • Total data sovereignty — your Cloudflare account, your keys
  • Strictly regulated environments — public sector, defence
  • Zero monthly cost — you pay Cloudflare Workers directly
  • Full code ownership — audit, fork, and contribute
  • PII_STRATEGY=edge for fully local inference
View on GitHub →
Managed Cloud Recommended for most teams
  • Zero ops — deploy in minutes, no wrangler needed
  • Managed audit logs, DPA included, automatic updates
  • Dashboard with PII detection metrics from request one
  • Dedicated support and SLA on Enterprise
  • Recommended for startups and regulated businesses
See pricing →
How to deploy

Four steps.
Your worker, live.

01
Fork the repo on GitHub
Go to github.com/privedge and fork the repository into your own account.
02
Install and deploy
Run npm install then wrangler deploy. Your worker is live in under two minutes.
03
Configure secrets
Add the PRIVEDGE_KEYS KV namespace and set CLOUD_API_KEY as a Wrangler secret.
04
Optional: enable edge inference
Set PII_STRATEGY=edge in wrangler.toml to run inference locally — no data ever leaves the node.
wrangler.toml

Edge inference config.
Two variables.

Set PII_STRATEGY to control where inference runs. Add EDGE_MODEL to choose the Workers AI model.

Available Workers AI models
@cf/meta/llama-3.2-3b-instruct Fast

High-volume tasks, low latency. Best choice for classification and extraction.

@cf/meta/llama-3.1-8b-instruct Balanced

General-purpose Q&A and summarization. Good quality-cost trade-off.

@cf/meta/llama-3.3-70b-instruct-fp8-fast Max quality

Complex reasoning, equivalent to GPT-3.5. Highest quality on the edge.

wrangler.toml 
[vars]
PII_STRATEGY = "edge"                                      # "anonymize" | "edge"
EDGE_MODEL   = "@cf/meta/llama-3.3-70b-instruct-fp8-fast" # Workers AI model
Architecture guarantees

Three facts.
Not promises.

01
Token map lives only in V8 heap

The PII substitution map is created at request time inside the isolate heap and destroyed with it. It is never written to any storage — KV, R2, or disk.

02
In edge mode, prompts never leave the node

When PII_STRATEGY=edge, Cloudflare Workers AI runs on the same node that received your request. No cloud API call, no external network hop, no third-party model.

03
MIT license — read every line

The worker code is MIT open source. You can audit, fork, modify, and redistribute it without restriction. No black box, no trust required.

Ready?

Start with the source.
Or skip the ops.

View source on GitHub Prefer zero ops? Try managed cloud →