Aiproxy
Local reverse proxy that redacts names and secrets before prompts reach LLM APIs
- Pricing
- Open source
- Platforms
- CLI, API, self-hosted
- Listing updated
- today
Similar tools
About Aiproxy
Aiproxy is an open-source AI tool for developer tools. It is available on CLI, API, self-hosted, macOS, Linux and Windows.
The README describes aiproxy as an anonymizing reverse proxy placed between an AI client, such as Claude Code, and the upstream provider. Requests are scrubbed of people's names, company names and other sensitive values by a local language model running through Ollama, then forwarded with placeholder tokens. When the reply returns, the originals are put back, so the user sees normal output while the provider only sees tokens.
The pipeline is built from pluggable processors. Bundled ones include a regex-based secrets filter, a local-LLM entity recognizer, an image redactor for scanned or embedded pictures, and a logger. Reversible mappings are kept in an encrypted SQLite or Postgres store, and a /dryrun endpoint lets you preview what would be redacted before relying on it.
The project targets individuals and organizations that need to keep customer data, client names or NDA material away from hosted LLM providers. The README is explicit that detection is not guaranteed and that operators must validate redaction for their own data. The software is licensed under PolyForm Noncommercial, so commercial use requires a separate license.
What it does
- Reversible token redaction
- Sensitive values are swapped for tokens before forwarding and restored in the response, with mappings kept in a local encrypted store.
- Local LLM entity detection
- An Ollama-hosted model detects people and company names, and deterministic allowlists, denylists and a name gazetteer back it up.
- Document and image redaction
- Handles images via OCR masking, PDFs and Office files such as docx, pptx and xlsx.
- Pluggable processor interface
- Custom filters can be added through a Processor interface without changing core code.
- Dry-run preview
- A /dryrun endpoint shows which redactions would be applied before live traffic is sent.
- Org deployment options
- Supports per-user isolation, Postgres, config hot-reload and a readiness endpoint for shared services.
Good for
- Using Claude Code on client projects without sending client names upstream
- Stripping customer PII from documents before an LLM analyzes them
- Keeping secrets and credentials out of prompts sent to a hosted provider
- Running a shared, centrally configured redaction gateway for a team
Pricing
Open source
Free for noncommercial use under PolyForm Noncommercial; commercial license available on request.
Aiproxy FAQ
Which providers does aiproxy support?
The README lists Anthropic /v1/messages and OpenAI /v1/chat/completions, with streaming and non-streaming responses.
Does aiproxy store my API key?
No. The README says the proxy holds no key and forwards the caller's x-api-key or Authorization bearer header.
Is it guaranteed to catch every sensitive value?
No. The README states that a local LLM can miss an entity, so detection is not guaranteed and redaction must be validated.
What license is it under?
The README states the PolyForm Noncommercial License 1.0.0, so noncommercial use is allowed and commercial rights are reserved.
Is Aiproxy free?
Yes. Aiproxy is open source, so you can use and self-host it for free.
What are the best alternatives to Aiproxy?
The closest alternatives to Aiproxy are Llmtee, ruby_llm, archestra, agentshield and codesnippets.
Where this comes from
- Description: written from the project's README.

