hackquest logo

Lemma

Save AI agent usage with the Lemma MCP server, which turns proven integration work into something agents can safely reuse.

비디오

기술 스택

Python
Solidity
Rust
Foundry
Stylus
Web3
Ethers
Node

설명

Lemma

Save AI agent usage with the Lemma MCP server that turns proven integration work into something agents can safely reuse.

Coding agents repeatedly spend tokens researching, rebuilding, and debugging integrations that have already been solved. Lemma checks whether reviewed integration work already fits the repository an agent is working on before it rebuilds the same thing from scratch.

How it works
• The agent requests an integration through the Lemma MCP server.
• Lemma profiles the repository locally and compares it against reviewed releases.
• It returns a deterministic decision: Reuse, Adapt, Build, or Decline.
• If a release matches, the agent can purchase the resolution through x402.
• The integration is recovered and applied locally.
• A predefined acceptance test verifies whether it actually works.
• The outcome is recorded as Passed, Failed, or Abandoned.

What is being verified
Every release can include the supported repository profile, exact integration bundle, dependency requirements, provenance, acceptance criteria, supporting evidence, pricing, and warranty terms.

The value is not hiding the source code. The value is answering one question reliably:

Does this proven integration actually apply to this repository?

Proof of the protocol
The core Lemma flow has already been exercised end to end on Arbitrum Sepolia.
• x402 payments were settled using USDC.
• A purchased resolution was applied and recorded as Passed.
• A separate purchased resolution was recorded as Failed.
• Provider funds were bonded against the resolution as warranty coverage.
• The failed resolution generated buyer credit.
• The buyer successfully claimed the testnet USDC refund.
• Compatibility outcomes were recorded through the deployed Stylus compatibility engine.

This tests both sides of the system: what happens when reused work succeeds, and what happens when it fails.

Why the warranty matters
Compatibility claims have consequences. Providers can place funds behind the integrations they publish. Successful outcomes strengthen compatibility history, while eligible verified failures can create buyer recourse from the provider-funded warranty.

Economic checks
Lemma is designed to avoid paying for reuse when rebuilding would be cheaper. A resolution is evaluated against the expected cost of rebuilding the integration, and releases without sufficient benchmark evidence can remain preview-only instead of being sold as proven savings.

The current testnet deployment proves the payment, compatibility, application, verification, warranty, failure, and refund mechanics. Measured production savings and a broader catalogue of verified integrations are the next step.

Future work

  • Agent reputation on ERC-8004. Every pass and failure becomes public feedback on the provider's agent, so any agent can check a track record before it buys.

  • Pay only after tests pass. The payment waits in escrow until the tests pass.

  • Outside providers. Any developer can publish a tested integration with a bond and earn on every reuse.

  • Measured benchmarks for every release, so prices come from real savings.

  • Independent reviewers instead of our team, and a shared multisig wallet instead of one owner key.

  • A security audit, then launch on Arbitrum One.

    Limitations

  • Early catalog. There is one first-party provider with two integrations, both for TypeScript MCP servers on Node. A release goes on sale only after a benchmark measures what it saves. The 1 October purchases used a demo release.

  • Trusted roles. Our team confirms failed tests before a refund is paid, and one owner key controls the contracts.

  • Test sandbox on Linux only. On macOS, a release's tests run without the sandbox.

  • Reputation not live yet. ERC-8004 agent reputation is built and tested on a local chain but not yet switched on for Arbitrum Sepolia.

해커톤 진행 상황

With the idea dating back to when my llm kept spending loads of token on working on reimplementations, every line of code was written during the Open House buildathon period. Week 1: shared schemas, the catalog and the deterministic matcher, the MCP bridge, the free preview, and the x402 purchase with the separate signer. Week 2: the warranty registry and the Stylus score engine the warranty pipeline (activation, results, refunds) the ERC-8004 reputation feed the dashboard and one-click install for five editors 1 Oct: contracts deployed on Arbitrum Sepolia. An agent made two real purchases there; one passed, one failed and was refunded on chain. Every step can be checked on Arbiscan. Since then: the hosted server, fixes found in the live run (event dates now come from the block itself), and a read-only check of the deployment that passes every check against the chain.

자금 모금 상태

Once the evidence cost and the revenue model is controlled (about 4-5 weeks of research and work) and established, the fundraising team will forward the pitch to known college alumni of IIT-Roorkee alert about the product achievements.
팀 리더
AAwaan Siddiqui
프로젝트 링크
부문
AIDeFiOther