Agentmark is a cryptographic provenance system designed to verify that code has been generated and committed exclusively by autonomous AI pipelines, without direct human intervention. It addresses the need for auditability and traceability in environments where AI-generated code is prevalent, providing a cryptographic audit trail for every commit. The tool is positioned for use cases such as agent-only open source projects, enterprise AI governance—especially in regulated industries or for compliance with legislation like the EU AI Act Article 50—and defense supply chains requiring procurement-grade verification of code provenance.
The core mechanism of agentmark is a chain of independently verifiable proofs embedded in each commit. These include an output hash that ensures the committed code matches the raw response from the language model byte-for-byte, a challenge token embedded in the task and echoed by the model to prove it processed a specific request, a provider-issued request ID confirming a real API call occurred, and an Ed25519 signature from the registered pipeline to establish pipeline identity. 0, the presence and structure of the signature are checked, with full cryptographic verification against a public key registry planned for future releases. Agentmark documents its honest limits and explicitly describes failure modes, such as the inability to distinguish between an agent and a human who perfectly replicates the agent workflow, or the reduced trust level when using local models without provider-issued request IDs.
Agentmark can be integrated into existing workflows with minimal changes. It is available as a Python SDK, a GitHub App, and a GitHub Action. Integration involves installing the tool via pip, wrapping LLM calls, signing the output to build a manifest, and adding a CI gate for provenance verification. The platform currently supports providers such as Anthropic and OpenAI, with plans for expanded verification capabilities and hardware attestation in future versions.
0 license and is developed by CloudDon Research. Agentmark is positioned as an open standard for cryptographic proof of AI-generated code provenance.
agentmark sits in PulseGate's AI & ML category. It focuses on verifying the authenticity and origin of AI-generated code to ensure trust and accountability. agentmark is an open-source project aimed at developers. The project is open source (Apache-2.0). agentmark is available on the web and the command line.
It is developed by agentmark-dev, and the product first shipped in 2026. Across PulseGate's embedding index, agentmark has few near neighbours, marking it as relatively distinct. Among its 5 catalogued features are provenance verification, AI code tracking, and cryptographic signatures.
Latest indexed changes and source events
agentmark.dev discovered by the PulseGate indexer
Other apps tracked under the same category.
