Tokenizer is an HTTP proxy that injects third party authentication credentials into requests. Clients encrypt third party secrets using the proxy's public key. When the client wants to send a request to the third party service, it does so via the proxy, sending along the encrypted secret in the Proxy-Tokenizer header. The proxy decrypts the secret and injects it into the client's request. To ensure that encrypted secrets can only be used by authorized clients, the encrypted data also includes instructions on authenticating the client.

Here's an example secret that the client encrypts using the proxy's public key:

secret = {
    inject_processor: {
        token: "my-stripe-api-token"
    },
    bearer_auth: {
        digest: Digest::SHA256.base64digest('trustno1')
    }
}

seal_key = ENV["TOKENIZER_PUBLIC_KEY"]
sealed_secret = RbNaCl::Boxes::Sealed.new(seal_key).box(secret.to_json)

The client configures their HTTP library to use the tokenizer service as it's HTTP proxy:

conn = Faraday.new(
    proxy: "http:https://tokenizer.flycast", 
    headers: {
        proxy_tokenizer: Base64.encode64(sealed_secret),
        proxy_authorization: "Bearer trustno1"
    }
)

conn.get("http:https://api.stripe.com")

The request will get rewritten to look like this:

GET / HTTP/1.1
Host: api.stripe.com
Authorization: Bearer my-stripe-api-token

Notice that the client's request is to http:https://api.stripe.com. In order for the proxy to be able to inject credentials into requests we need to speak plain HTTP to the proxy server, not HTTPS. The proxy transparently switches to HTTPS for connections to upstream services. This assumes communication between the client and tokenizer happens over a secure transport (a VPN).

The processor dictates how the encrypted secret gets turned into a credential and added to the request. The example above uses inject_processor, which simply injects the verbatim secret into a request header. By default, this injects the secret into the Authorization: Bearer header without further processing. The inject_processor can optionally specify a destination and/or printf-style format string to be applied to the injection of the credential:

secret = {
    inject_processor: {
        token: "my-stripe-api-token",
        dst:   "X-Stripe-Token",
        fmt:   "token=%s",
    },
    bearer_auth: {
        digest: Digest::SHA256.base64digest('trustno1')
    }
}

This will result in the header getting injected like this:

X-Stripe-Token: token=my-stripe-api-key

Aside from inject_processor, we also have inject_hmac_processor. This creates an HMAC signatures using the key stored in the encrypted secret and injects that into a request header. The hash algorithm can be specified in the secret under the key hash and defaults to SHA256. This processor signs the verbatim request body by default, but can sign custom messages specified in the msg parameter in the Proxy-Tokenizer header (see about parameters bellow). This processor also respects the dst and fmt options.

secret = {
    inject_hmac_processor: {
        key: "my signing key",
        hash: "sha256"
    },
    bearer_auth: {
        digest: Digest::SHA256.base64digest('trustno1')
    }
}

If the destination/formatting might vary between requests, inject_processor and inject_hmac_processor can specify an allowlist of dst/fmt parameters that the client can specify at request time. These parameters are supplied as JSON in the Proxy-Tokenizer header after the encrypted secret.

secret = {
    inject_processor: {
        token: "my-stripe-api-token"
        allowed_dst: ["X-Stripe-Token", "Authorization"],
        allowed_fmt: ["Bearer %s", "token=%s"],
    },
    bearer_auth: {
        digest: Digest::SHA256.base64digest('trustno1')
    }
}

seal_key = ENV["TOKENIZER_PUBLIC_KEY"]
sealed_secret = RbNaCl::Boxes::Sealed.new(seal_key).box(secret.to_json)

processor_params = {
    dst: "X-Stripe-Token", 
    fmt: "token=%s"
}

conn.headers[:proxy_tokenizer] = "#{Base64.encode64(sealed_secret)}; #{processor_params.to_json}"

conn.get("http:https://api.stripe.com")

If a client is fully compromised, the attacker could send encrypted secrets via tokenizer to a service that simply echoes back the request. This way, the attacker could learn the plaintext value of the secret. To mitigate against this, secrets can specify which hosts they may be used against.

secret = {
    inject_processor: {
        token: "my-stripe-api-token"
    },
    bearer_auth: {
        digest: Digest::SHA256.base64digest('trustno1')
    },
    allowed_hosts: ["api.stripe.com"],
    # or
    # allowed_host_pattern: ".*\.stripe\.com$"
}

Assuming you have flyctl installed, start by cloning this repository

git clone https://github.com/superfly/tokenizer
cd ./tokenizer

create a fly.io app:

fly app create
export FLY_APP="<name of app>"

generate a private (open) key:

OPEN_KEY=$(openssl rand -hex 32)
fly secrets set --stage OPEN_KEY=$OPEN_KEY

Deploy the app without making it available on the internet¹:

fly deploy --no-public-ips

Tokenizer is now deployed and accessible to other apps in your org at <name of app>.flycast. The deploy logs will contain the public (seal) key, which can be used for encrypting secrets.

¹Assigning a public IP address to the app is not recommended, since it will happily proxy traffic to private IP addresses. If you require a public deployment, consider running tokenizer in a separate, dedicated organization or using it in conjuction with smokescreen.

Tokenizer is totally stateless, so it's simple to deploy anywhere.

Assuming you have Golang installed, you can build and install tokenizer in /usr/local/bin by running

GOBIN=/usr/local/bin go install github.com/superfly/tokenizer/cmd/tokenizer@latest

Generate a private (open) key:

export OPEN_KEY=$(openssl rand -hex 32)

Run the tokenizer server:

tokenizer

The output will contain the public (seal) key, which can be used for encrypting secrets.

See the READMEs in github.com/superfly/tokenizer/cmd/tokenizer and github.com/superfly/tokenizer/cmd/curl for instructions on running/testing tokenizer locally.

Tokenizer is configured with the following environment variables:

• OPEN_KEY - The hex encoded 32 byte private key is used for decrypting secrets.
• LISTEN_ADDRESS - The address (ip:port) to listen on.
• FILTERED_HEADERS - A comma separated list of request headers to strip from client requests.