Enclavia

EnclaviaProvable computation, as simple as pushing a Docker image.

Run your container inside an attested enclave. End-to-end encryption from the browser. Public beta.

Install the CLI
Run a sample

One command to deploy

`enclavia deploy myapp:v1` creates the enclave, pushes your image, and follows the build live until the enclave is running inside AWS Nitro. The individual `create` and `push` steps remain available for scripts and agents.

Attested by default

Every connection performs a Noise NN handshake, requests an attestation document, and verifies PCR0/1/2 against the image you pushed. No remote agent, no extra steps.

Reproducible builds

`enclavia reproduce` rebuilds the enclave image from the exact recorded sources and compares the PCRs to the deployed enclave, so anyone can verify the running code matches what was pushed, not just trust the attestation. Provable, end to end.

Encrypted from the browser

The client SDK ships for Rust, for browsers and Node as `@enclavia/client-wasm` on npm, and for Dart/Flutter as `enclavia_dart` on pub.dev. It speaks the Noise channel directly to the enclave over a WebSocket proxy, so the plaintext never leaves the user's device or the enclave.

Storage sealed to your enclave

Optional persistent volumes are LUKS-encrypted with a passphrase held in AWS KMS. The key is only released after attestation matches your image's PCRs, so the data is bound to the same identity as the code that wrote it.

Drive it from your AI agent

Two ways to let an agent manage your enclaves in natural language. A hosted MCP server for any MCP-aware client (Claude, ChatGPT, Cursor, the OpenAI Codex CLI), or the CLI plus an agent skill for local agents that already have a shell, which is more token-efficient. Both use the same OAuth identity as the CLI, scoped per user and revocable from the dashboard.

What is Enclavia ​

Enclavia is a managed platform for running container workloads inside hardware-attested enclaves. You point it at a Docker image; it builds an enclave image, boots it on Nitro hardware, and exposes it behind a WebSocket proxy that speaks an end-to-end encrypted channel directly to the enclave.

The pieces a user touches:

  • enclavia CLI — authenticate, push images, create and manage enclaves. On crates.io as enclavia-cli.
  • enclavia client SDK — connect from a server, browser, or app, verify attestation, send HTTP through the encrypted channel. On crates.io as enclavia (Rust), on npm as @enclavia/client-wasm (browsers, Node 22+, Deno), and on pub.dev as enclavia_dart (Dart, Flutter).
  • Backend API — https://api.beta.enclavia.io. Not a supported public interface during the beta: the CLI, the MCP server, and the agent skill are the supported surfaces on top of it. A standalone REST reference may be published once the API stabilizes.
  • MCP server — https://mcp.beta.enclavia.io/mcp. Lets any MCP-aware agent (Claude, ChatGPT, Cursor, Codex, …) drive your enclaves with the same identity the CLI uses.

Where to start ​

The fastest path to seeing Enclavia work is to run a sample app end to end. For your own workload the steps are:

  1. Install the CLI.
  2. Authenticate by approving a session in the web UI.
  3. Deploy your image with enclavia deploy myapp:v1, which creates the enclave, pushes the image, and follows the build until it's running. (Scripts and agents should run the individual create and push steps instead.)
  4. Connect to it from your code, or point an AI agent (Claude, ChatGPT, Cursor, Codex, …) at the same enclaves over MCP.

Beta scope ​

The public beta runs at beta.enclavia.io and is intended for evaluation. Image references resolve against registry.beta.enclavia.io under your handle (the user-chosen identifier set during onboarding). The CLI talks to https://api.beta.enclavia.io and the encrypted client connects to enclaves under enclaves.beta.enclavia.io.

For AI agents ​

A machine-readable index of these docs is published at /llms.txt — the convention for surfacing documentation to LLMs without parsing HTML. The complete corpus, every page rendered into a single file, is at /llms-full.txt. Feed either to your agent of choice.

Built for AI agents too — fetch /llms.txt for a machine-readable index of these docs, or /llms-full.txt for the full corpus in one file.

Enclavia · provable computation, as simple as pushing a Docker image.