GitHub

Open source · MIT licence · Pre-release

Short-lived secrets for applications and AI agents

Simault is a small secrets service you run yourself. Each registered application gets one secret that expires after an hour, encrypted before it ever reaches the database.

It is for platform teams and agent builders who want to hand out credentials that stop working on their own.

Get started View on GitHub 

$ curl https://vault.example.com/api/secrets/payment-service \
    -H "X-AGENT-TOKEN: $ACCESS_TOKEN"

{
  "appId": "payment-service",
  "secretValue": "Xy9zQ1p0...",
  "createdAt": "2026-11-11T12:00:00Z"
}

An application reads its current secret with a one-hour access token.

How it works

An administrator decides which applications exist. Each application proves who it is with its own RSA key. Nobody shares a long-lived password.

  1. Register

    An administrator registers the application and receives a one-time registration token.

  2. Upload a public key

    The application creates an RSA key pair and completes registration with its public key. The private key never leaves it.

  3. Sign for a token

    The application signs its ID and the current time, and exchanges the signature for a one-hour access token. Each signed request works once.

  4. Read the secret

    With the token, the application reads or rotates its own secret. When the secret expires, the next read returns a new one.

Registration and secret retrieval flow The administrator registers an application with Simault. The agent uploads its public key, exchanges a signed request for an access token, and reads its secret. Simault stores only encrypted data in MongoDB. Administrator Application or AI agent Simault encrypts and decrypts MongoDB ciphertext only 1. register 2. public key 3. token 4. secret encrypted fields

What it does

Encryption

Secrets and public keys are encrypted in the application with MongoDB client-side field level encryption. The database stores only ciphertext, so a database administrator or a stolen backup does not reveal them.

Short-lived by default

Secrets and access tokens last one hour unless you change it. Removing an application deletes its tokens and its secret immediately.

Access control

Policies limit what each application may do: read or rotate, and from which network ranges. Token, admin and secret endpoints are rate limited per client.

Dynamic credentials

An optional engine makes each secret a temporary MongoDB database user, dropped on rotation, expiry or revocation.

Built for AI agents

An MCP server exposes six tools to AI clients, and two A2A v1.0 agents publish agent cards with a JSON-RPC endpoint. How agents connect.

Runs on your Kubernetes

A Helm chart covers Amazon EKS, Google GKE, Azure AKS and Red Hat OpenShift, with a written guide for each. Deployment options.

Project status

Simault is pre-release software, version 1.0.0-SNAPSHOT. Read this before you rely on it.

  • It has had no independent security review.
  • The tested configuration uses a local master key file. Cloud KMS support for AWS, Google Cloud, Azure and KMIP is implemented but has not been tested against a real KMS.
  • The Helm chart passes lint and renders for each platform, but has not yet been installed on a real cluster.
  • There is no public container image. You build your own from the repository.
  • Simault serves plain HTTP. It must run behind an ingress or proxy that terminates TLS.

The full list is on the security page.