Primus Labs posts zkTLS AI demo

Primus Labs used a September 8 social-media push to show how it wants AI agents to prove they pulled real-world data without exposing the source data itself. The post linked to a zkTLS demo and described the system as a way to verify that an agent fetched information such as stock prices while keeping the underlying payload private, according to the X thread referenced in the briefing. Primus presents zkTLS as cryptographic plumbing for cases where an agent needs to show it queried a real website or API, rather than inventing an answer. Public Primus materials describe the product as part of a broader stack for “data verification and computation” built around zkTLS and other cryptographic tools. ### How is Primus describing zkTLS in this demo? Primus’ public tutorial says zkTLS “verifies TLS data authenticity while preserving privacy, without modifying servers.” The same tutorial says the system is meant to enable secure use of Web2 data and that developers can build applications around off-chain data using templates and SDKs. The September 8 thread described in the briefing framed that pitch around AI agents. (github.com) In that formulation, the proof is not that the data becomes public; it is that the agent can later show it fetched authentic data from a real endpoint. The examples cited in the post included stock-price checks for DeFi trading bots and enterprise compliance workflows, according to the briefing’s account of the thread. (github.com) ### What problem is this supposed to solve for AI agents? Primus’ GitHub materials say one use case involves an “agent program” receiving user credentials and generating zkTLS attestations on the user’s behalf. That is a more specific claim than the social post: it suggests Primus is targeting agent workflows where software, not a human clicking through a browser, performs the fetch and proof generation. (github.com) The same repository says end users can also generate attestations themselves and share them on-chain. That gives Primus two operating modes in its own documentation — direct user-side proof generation and backend or agent-side generation — both aimed at producing attestations that can be checked later. ### What does Primus say developers actually get? Primus’ tutorial says the developer stack includes pre-defined data templates and zkTLS SDKs. (github.com) The company says developers can select a template, use a template ID, and interact with a decentralized attestor group it calls Primus AlphaNet. A separate Primus SDK repository says developers can verify data through web endpoint responses and then validate a Primus signature before triggering business logic. (github.com) That repository also says private-data access may require an authorized token and that the SDK supports “Proxy TLS” and “MPC TLS” models. ### How would a proof get checked after the fetch? Primus’ extension-core documentation lays out a five-step flow: create a project, configure request and response parameters, execute the zkTLS protocol, verify the result, and then run the business logic. (github.com) In Primus’ description, the verification step includes validating a Primus signature tied to the attestation. Primus has also published verifier tooling for blockchain use. (github.com) Its public repositories include zkTLS smart-contract code and an Aztec-focused verifier package that the company says can verify Primus zkTLS attestations in smart contracts. ### What is public today beyond the social post? Primus’ GitHub organization lists multiple zkTLS-related repositories, including a JavaScript SDK, extension components and contract libraries, with several repositories updated in mid-2026. (github.com) The company’s GitHub profile describes Primus as infrastructure for validating web data and processing it with privacy guarantees. The public materials available now are the company website, the tutorial repository, the SDK repositories and verifier code. (github.com) Those are the places where developers can inspect the mechanics behind the September 8 demo and compare the AI-agent pitch in the X thread with the implementation details Primus has already published. (primuslabs.xyz) (github.com)