Aegis is an open-source, self-hosted AI governance and evidence gateway. It applies request controls, bounded streaming PII redaction, durable audit records, and client-verifiable MMR inclusion proofs to supported LLM routes. Python clients and TypeScript helpers ship as aegis-latent-sdk 4.0.0. GitHub’s Release label is v4.0.1; source and attached package assets remain 4.0.0.
Hi Product Hunt,
I built Aegis Latent Core for teams that need more than a conventional proxy log around LLM traffic. It is a self-hosted AI governance and evidence gateway for supported OpenAI-compatible routes and, when the gateway is configured for it, native Anthropic Messages traffic.
For governed non-streaming outcomes, Aegis hashes and appends evidence to its authoritative JSONL WAL before returning the response. For admitted SSE streams, it incrementally emits sanitized events through a bounded queue and commits one exact-byte terminal summary before the terminal marker. It can generate portable Merkle Mountain Range inclusion proofs, which the Python and TypeScript SDKs can verify against an independently provisioned trusted root. A valid proof establishes inclusion relative to that root; it does not by itself establish external immutability, non-repudiation, or legal provenance.
The repository also includes:
• Opt-in, bounded streaming redaction for supported text patterns
• Typed Python OpenAI and Anthropic client subclasses within tested boundaries
• TypeScript gateway helpers and Web Crypto proof verification
• A read-only dashboard and bounded forensic export workflow
• Finite-state artifacts checked with Z3, Lean 4, and TLA+/TLC; these checks are not a refinement proof of the full implementation
Install the published 4.0.0 SDKs:
python -m pip install 'aegis-latent-sdk[openai]==4.0.0'
npm install aegis-latent-sdk@4.0.0 openai@^6.49.0
Version note: GitHub labels the current Release v4.0.1. Its ref is a lightweight commit ref, not a signed tag, and the associated tag-triggered release and publication workflows did not complete successfully. The source version anchors, ten attached asset names, and public PyPI/npm packages identify version 4.0.0.
Repository: https://github.com/JuanLunaIA/ae...
GitHub Release label: https://github.com/JuanLunaIA/ae...
Python 4.0.0: https://pypi.org/project/aegis-l...
TypeScript 4.0.0: https://www.npmjs.com/package/ae...
The repository describes AGPLv3 community use and separate commercial terms, but no support, SLA, certification, or production acceptance is implied. I would especially value feedback on the streaming evidence boundary, independently trusted MMR roots, and SDK ergonomics.