ProveKit

A modular zero-knowledge proof toolkit optimized for mobile devices.

Requirements

This project makes use of Noir's nargo to compile circuits and generate test artifacts. Make sure to walk through the Quick Start section to install the noir toolchain. Note that we require a specific version of the toolchain, so make sure to override the version with the following command.

noirup --version v1.0.0-beta.11

Demo instructions

NOTE: The example below is being run for single example poseidon-rounds. You can use different example to run same commands.

Compile the Noir circuit:

cd noir-examples/poseidon-rounds
nargo compile

Prepare the Noir program (generates prover and verifier files):

cargo run --release --bin provekit-cli prepare ./target/basic.json --pkp ./prover.pkp --pkv ./verifier.pkv

Generate the Noir Proof using the input Toml:

cargo run --release --bin provekit-cli prove ./prover.pkp ./Prover.toml -o ./proof.np

Verify the Noir Proof:

cargo run --release --bin provekit-cli verify ./verifier.pkv ./proof.np

Generate inputs for Gnark circuit:

cargo run --release --bin provekit-cli generate-gnark-inputs ./prover.pkp ./proof.np

Recursively verify in a Gnark proof (reads the proof from ../ProveKit/prover/proof):

cd ../../recursive-verifier
go run .

Benchmarking

Benchmark against Barretenberg:

Note: You can install Barretenberg from here.

Note: You can install hyperfine using brew on OSX: brew install hyperfine.

cd noir-examples/poseidon-rounds
cargo run --release --bin provekit-cli prepare ./target/basic.json --pkp ./prover.pkp --pkv ./verifier.pkv
hyperfine 'nargo execute && bb prove -b ./target/basic.json -w ./target/basic.gz -o ./target' '../../target/release/provekit-cli prove ./prover.pkp ./Prover.toml'

Profiling

Custom built-in profile (Memory usage)

The provekit-cli application has written custom memory profiler that prints basic info about memory usage when application runs. To run binary with profiling enabled run it with cargo --features profiling param or compile with it.

cargo run --release --bin provekit-cli --features profiling prove ./prover.pkp ./Prover.toml -o ./proof.np

Using tracy (CPU and Memory usage)

Tracy tool website. To install tracy tool on OSX use brew: brew install tracy.

Important: integration is done with Tracy Profiler 0.11.1. It is newest version available from brew. Newer version may require updating dependencies as tracy is using its own protocol between app and tracy tool that changes with each major version.

TLDR; Tracy is an interactive tool to profile application. There is integration plugin for rust that works with standard tracing annotation. For now it is integrated into provekit-cli binary only. Collecting profiling data requires tracy to run during application profiling. You may noticed that it makes application to run much longer but mostly due to data transfer between the application and the tracy running along.

Usage:

1. Start tracy from command line

tracy

2. Leave all fields with defaults and just click Connect button. It will cause tracy to start listening on the localhost for incoming data.
3. Compile noir-r1cs-profiled binary.

cargo build --release --bin provekit-cli --features profiling

4. (OSX only) If you want to check call stacks additional command needs to be run (base on tracy instruction). The command must be run against each binary that is being profiled by tracy. This will create directory next to the binary provided with .dSYM suffix (ex. ../../target/profiled-cli.dSYM). Directory will contain the debug symbols and paths extracted with different format that is compatible with tracy tool. It must be rerun after each changes made to provekit-cli app.

 dsymutil ../../target/release/provekit-cli

5. Now start the application to profile:

../../target/release/provekit-cli prove ./prover.pkp ./Prover.toml -o ./proof.np

6. Go back to tracy tool. You should see that it receives data. App is interactive.

Using samply (CPU usage)

Samply tool website with instructions to install. It will start local server and open a webpage with interactive app to view results. This does not require to run binary with profiling enabled.

samply record -r 10000 -- ./../../target/release/provekit-cli prove ./prover.pkp ./Prover.toml -o ./proof.np

Using instruments (Memory usage) - OSX only

Cargo instruments tool website with instructions to install. It will open results using built-in Instruments app. Results are interactive.

cargo instruments --template Allocations --release --bin provekit-cli prove ./prover.pkp ./Prover.toml -o ./proof.np

Samply tool website with instructions to install. It will start local server and open a webpage with interactive app to view results. This does not require to run binary with profiling enabled.

samply record -r 10000 -- ./../../target/release/provekit-cli prove ./prover.pkp ./Prover.toml -o ./proof.np

Benchmarking

Run the benchmark suite:

cargo test -p provekit-bench --bench bench

Architecture

ProveKit follows a modular architecture with clear separation of concerns:

Core Modules

• provekit/common/ - Shared utilities, core types, and R1CS abstractions
• provekit/r1cs-compiler/ - R1CS compilation logic and Noir integration
• provekit/prover/ - Proving functionality with witness generation
• provekit/verifier/ - Verification functionality

Tooling

• tooling/cli/ - Command-line interface (provekit-cli)
• tooling/provekit-bench/ - Benchmarking infrastructure
• tooling/provekit-gnark/ - Gnark integration utilities

High-Performance Components

• skyscraper/ - Optimized field arithmetic for M31/CM31 fields
• playground/ - Research and experimental implementations

Examples & Tests

• noir-examples/ - Example circuits and test programs
• gnark-whir/ - Go-based recursive verification using Gnark

Dependencies

This project depends on the following libraries, which are developed in lockstep:

• 🌪️ WHIR
• Spongefish
• gnark-skyscraper
• recursive-verifier
• noir