Skip to content

MoonZagallo/witnex

 
 

Witnex

CI License: MIT OR Apache-2.0

Verifiable AI agent framework — every action produces a cryptographic commitment + ZK proof, verifiable by a third party without re-running it.

Witnex records each agent execution (LLM call, tool use, decision) as a tamper-evident execution trace and proves that trace well-formed with a Risc0 ZK proof. A verifier confirms what the agent did without access to the prover, the LLM, or the original plaintext.

What Witnex proves (Phase 1)

  • Input/output integrity — the agent committed to specific inputs and outputs and cannot tamper with them after the fact.
  • Execution trace integrity — tool calls happened in the claimed order with the claimed parameters.

Not in Phase 1: Witnex does not prove the LLM inference itself was correct. That is the harder zkML problem, intentionally deferred. See docs/architecture.md.

Repository layout

witnex/
├── crates/                 # Rust workspace
│   ├── witnex-core/        # Agent runtime + trace types (Agent, ExecutionTrace, ToolCall)
│   ├── witnex-prover/      # Risc0 guest + host (proof generation)
│   ├── witnex-verifier/    # Standalone verifier
│   └── witnex-cli/         # Demo CLI (`witnex`)
├── packages/               # TypeScript / pnpm workspace
│   ├── sdk/                # @witnex/sdk
│   └── examples/           # Demo apps
└── docs/                   # Architecture, positioning, ADRs

Status

Phase 1, Prompt 1 complete — workspace scaffold, core types/traits, and the architecture document. No proving or LLM logic is implemented yet; that is the next slice (witnex demo summarize / witnex verify).

Prerequisites

  • Rust ≥ 1.85 (edition 2024) — install via rustup. On Windows the default msvc target needs the Visual C++ Build Tools; the rustup x86_64-pc-windows-gnu toolchain is a self-contained alternative.
  • Node ≥ 22.13 (required by pnpm 11.7+) and pnpm ≥ 11.

Build

# Rust workspace
cargo build

# TypeScript workspace
pnpm install
pnpm -r build

Contributing

See CONTRIBUTING.md for dev setup, the PR process, and code style. All participation is governed by our Code of Conduct.

License

Dual-licensed under either of

at your option.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

About

Verifiable AI agent framework — cryptographic commitments + ZK proofs for agent execution traces.

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Rust 93.2%
  • TypeScript 5.7%
  • JavaScript 1.1%