CLI

​

Overview

The Tinfoil CLI provides a command-line interface for making verified HTTP requests to Tinfoil enclaves and validating attestation documents. It supports all major AI inference operations including chat completions, audio transcription, and text embeddings.

​

Installation

​

Pre-built Binaries

Download the latest release for your OS from the Releases page.

​

Install Script

You can install tinfoil CLI using our install script. This script automatically detects your operating system and architecture, downloads the correct binary, and installs it to /usr/local/bin.

curl -fsSL https://github.com/tinfoilsh/tinfoil-cli/raw/main/install.sh | sh

Note: If you receive permission errors (for example, if you’re not running as root), you may need to run the command with sudo:

curl -fsSL https://github.com/tinfoilsh/tinfoil-cli/raw/main/install.sh | sh

​

Build from Source

1. Ensure you have Go installed.
2. Clone the repository:

git clone https://github.com/tinfoilsh/tinfoil-cli.git
cd tinfoil-cli

3. Build the binary:

go build -o tinfoil

​

Command Reference

Usage:
  tinfoil [command]

Available Commands:
  attestation  Attestation commands (verify or audit)
  chat         Chat with a language model
  audio        Transcribe audio files using Whisper
  embed        Generate text embeddings
  proxy        Run a local HTTP proxy
  completion   Generate the autocompletion script for the specified shell
  help         Help about any command
  http         Make verified HTTP requests

Flags:
  -e, --host string           Enclave hostname
  -h, --help                  help for tinfoil
  -r, --repo string           Source repo

Use "tinfoil [command] --help" for more information about a command.

​

Usage Examples

​

Chat

Chat with AI models using natural language.

​

Basic Usage

tinfoil chat -m llama3-3-70b -k "YOUR_API_KEY" "Why is tinfoil now called aluminum foil?"

​

Streaming Example

tinfoil chat -m llama3-3-70b -k "YOUR_API_KEY" "Explain quantum computing in simple terms"

​

Specifying a Custom Model

For custom models not included in config.json, supply the model name along with the -e and -r overrides:

tinfoil chat -m custom-model -k "YOUR_API_KEY" "Explain string theory" \
  -e custom.enclave.example.com \
  -r cool-user/custom-model-repo

If you omit -e or -r for a model that isn’t in the configuration, a warning will be displayed prompting you to specify these flags.

Command Options:

• -m, --model: The model name to use for chat. Must be specified.
• -k, --api-key: The API key for authentication.
• -e, --host: The hostname of the enclave. Optional if defined in the config file.
• -r, --repo: The GitHub repository containing code measurements. Optional if defined in the config file.

​

Audio Transcription

The audio command allows you to transcribe audio files using Whisper. By default, it uses the whisper-large-v3-turbo model.

​

Basic Usage

tinfoil audio -k "YOUR_API_KEY" -f path/to/audio/file.mp3

This command uses the default model whisper-large-v3-turbo and loads the enclave host and repo values from config.json.

​

Specifying a Custom Model

tinfoil audio -m custom-whisper-model -k "YOUR_API_KEY" -f path/to/audio/file.mp3 \
  -e custom.enclave.example.com \
  -r cool-user/custom-model-repo

Command Options:

• -m, --model: The model name to use for transcription. Defaults to whisper-large-v3-turbo.
• -k, --api-key: The API key for authentication.
• -f, --file: The audio file to transcribe.
• -e, --host: The hostname of the enclave. Optional if defined in the config file.
• -r, --repo: The GitHub repository containing code measurements. Optional if defined in the config file.

​

Text-to-Speech

The CLI can generate audio from text using the Kokoro model via the HTTP command:

# Generate speech and save to file
tinfoil http POST \
  -e audio-processing.model.tinfoil.sh \
  -r tinfoilsh/confidential-audio-processing \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  https://audio-processing.model.tinfoil.sh/v1/audio/speech \
  -d '{
    "model": "kokoro",
    "voice": "af_sky+af_bella",
    "input": "Hello world! This is a test of text-to-speech synthesis.",
    "response_format": "mp3"
  }' > output.mp3

echo "Speech saved to output.mp3"

Available voices include af_sky, af_bella, af_heart, and combinations like af_sky+af_bella.

​

Text Embeddings

The embed command allows you to generate embeddings for text inputs. By default, it uses the nomic-embed-text model.

​

With Default Model

tinfoil embed -k "YOUR_API_KEY" "This is a text I want to get embeddings for."

This command uses the default model nomic-embed-text and loads the enclave host and repo values from config.json.

​

With Multiple Text Inputs

You can provide multiple text inputs to get embeddings for all of them:

tinfoil embed -k "YOUR_API_KEY" "First text" "Second text" "Third text"

​

Specifying a Custom Model

tinfoil embed -m custom-embed-model -k "YOUR_API_KEY" "Text to embed" \
  -e custom.enclave.example.com \
  -r cool-user/custom-model-repo

Command Options:

• -m, --model: The model name to use for embeddings. Defaults to nomic-embed-text.
• -k, --api-key: The API key for authentication.
• -e, --host: The hostname of the enclave. Optional if defined in the config file.
• -r, --repo: The GitHub repository containing code measurements. Optional if defined in the config file.

​

Attestation Verification

​

Verify Attestation

Use the attestation verify command to manually verify that an enclave is running the expected code. The output will be a series of INFO logs describing each verification step.

Sample successful output:

$ tinfoil attestation verify \
  -e llama3-3-70b-p.model.tinfoil.sh \
  -r tinfoilsh/confidential-llama3-3-70b-prod
INFO[0000] Fetching latest release for tinfoilsh/confidential-llama3-3-70b-prod 
INFO[0000] Fetching sigstore bundle from tinfoilsh/confidential-llama3-3-70b-prod for digest f2f48557c8b0c1b268f8d8673f380242ad8c4983fe9004c02a8688a89f94f333 
INFO[0001] Fetching trust root                          
INFO[0001] Verifying code measurements                  
INFO[0001] Fetching attestation doc from llama3-3-70b-p.model.tinfoil.sh 
INFO[0001] Verifying enclave measurements               
INFO[0001] Public key fingerprint: 5f6c24f54ed862c404a558aa3fa85b686b77263ceeda86131e7acd90e8af5db2 
INFO[0001] Remote public key fingerprint: 5f6c24f54ed862c404a558aa3fa85b686b77263ceeda86131e7acd90e8af5db2 
INFO[0001] Measurements match  

​

Audit Attestation

You can also verify attestations at random and record a machine-readable audit log. Use the attestation audit command for this purpose.

By default the audit record is printed to stdout as JSON. To write it to a file, use the -l/--log-file flag:

tinfoil attestation audit \
  -e llama3-3-70b-p.model.tinfoil.sh \
  -r tinfoilsh/confidential-llama3-3-70b-prod \
  -l /var/log/tinfoil_audit.log

The audit log record includes the timestamp, enclave host, code and enclave measurement fingerprints, and the verification status.

​

Proxy

Use tinfoil proxy to start a local HTTP proxy that verifies connections and forwards them to the specified enclave:

tinfoil proxy \
  -r tinfoilsh/confidential-llama3-3-70b-prod \
  -e llama3-3-70b-prod.model.tinfoil.sh \
  -p 8080