# Alkahest

> Private and uncensored LLM inference platform with end-to-end encryption and trusted execution environment deployment.

- Role: Private AI infrastructure
- Status: Public site and open source organization
- Stack: Trusted execution environments, end-to-end encryption, LLM inference, TypeScript
- Page: https://thomasjvu.com/projects/alkahest
- Markdown: https://thomasjvu.com/projects/alkahest.md

## Links

- [Site](https://alkahest.ai/)
- [GitHub](https://github.com/alkahest-ai)

## Metrics

- Privacy model: E2EE — Inference product direction centered on encrypted user interaction.
- Deployment: TEE — Infrastructure direction built around trusted execution environments.

## Writeup

## Overview

Alkahest is a private LLM inference product: encrypted interaction, TEE-backed deployment, and OpenAI-compatible routes instead of a generic hosted chat wrapper.

## Problem

Most hosted inference asks you to trust the operator with prompts. Alkahest is an attempt to make the privacy boundary visible — what is protected, what hardware is trusted, and where that trust ends.

## What I built

Product surface for private model interaction, attestation-aware UI, and an API-shaped landing that shows chat, image, transcript, speech, video, and model routes.

## Hard parts

Privacy copy is easy to fake. The hard part is making the trust model legible without turning the app into a cryptography lecture.

## Result

Public site: [alkahest.ai](https://alkahest.ai/). Org: [github.com/alkahest-ai](https://github.com/alkahest-ai). Pairs with Mystery Gift as TEE-adjacent infrastructure work.