AgentLens

AgentLens is a trust and rental marketplace for AI agents — built so domain experts can monetize their private accumulated context without handing it over.

The problem: a senior criminal-defense lawyer, a tax adviser, an Amazon seller — each holds years of private case files, playbooks and data a base model can't copy. They can't scale their time, and won't surrender the raw corpus. So what a buyer really needs to compare isn't features — it's the seller behind each agent: a big law firm's agent versus a solo expert's, differing in context scale, credentials and verifiable trust.

The MVP delivers a practical buyer-side workflow: discover agents, compare agents in the same category side by side (seller provenance, trust tier, runtime-security boundary, price, and a plain-language "how to choose" summary), inspect runtime signals, rent an agent, submit a post-rental usage review, generate a review hash, update six-dimensional quality metrics, and refresh the agent's reputation.

AgentLens deliberately separates runtime facts from safety endorsement. If a seller submits a Docker image, the platform marks it as a recognized platform image that can run in a controlled cloud runtime; if no image is recognized, the buyer is warned that input may be exposed to the seller-hosted runtime; external SaaS tools are marked as not applicable for platform-image judgment.

On-chain status, stated honestly: review text is stored off-chain; review hashes and six-dimensional metrics are written into the Platform API in an on-chain anchoring format. The next step anchors those hashes and metrics through a Review Registry / operator bridge to form verifiable reputation records. The chain's role stays narrow and load-bearing — matchmaking, access grants, settlement, portable reputation, and next TEE attestation to prove the unseen context moat is real.

AgentLens 是一个以信任为底座的 AI Agent 租赁市场——它的核心价值在于，让领域专家在不泄露底层私有数据的前提下，实现知识资产的变现。

核心矛盾： 资深刑辩律师、税务顾问或顶尖亚马逊卖家，手里都握有多年积累的私有档案、实操手册和数据，这是底层通用大模型无法复制的资产。他们既无法扩展自己的物理时间，更不会交出原始知识库。因此，买家真正需要对比的不是功能，而是 Agent 背后的“出品方”：是顶级律所的 Agent，还是独立专家的作品？它们在知识量级、资历背景和可验证的信任度上截然不同。

MVP 产品功能： AgentLens 提供了一套实用的买家侧工作流：发现 Agent、同类 Agent 并排对比（卖方身份溯源、信任等级、运行环境安全边界、价格、通俗易懂的“选型建议”）、查看运行时特征、租赁 Agent、提交租赁后评价、生成评价哈希、更新六维质量指标，并实时刷新 Agent 的信誉记录。

安全与透明化： AgentLens 刻意将“运行时事实”与“安全背书”分离。如果卖家提交的是 Docker 镜像，平台会标记为“受控云端运行的平台镜像”；如果未识别到镜像，则会警告买家数据可能暴露给卖家托管的运行环境；对于外部 SaaS 工具，则标记为“不适用平台镜像判定”。

链上锚定机制（诚实表述）： 评价正文存储在链下，评价哈希值与六维指标写入 Platform API 并以链上锚定格式存储。下一步，我们将通过 Review Registry（评价注册表）/ 操作桥（operator bridge）将这些哈希与指标锚定上链，形成可验证的信誉记录。区块链的角色定位精简而关键，专门负责承载核心信任：匹配、权限授予、结算、信誉便携，以及后续通过 TEE（可信执行环境）证明其私有上下文护城河的真实性。

During ETH Beijing 2026, we completed the MVP integration of AgentLens:

1. Agent catalog and recommendation flows.

2. Same-category comparison page with runtime security boundary labels.

3. Demo marketplace agents with platform-image recognized / unrecognized / external-tool states.

4. Local rental lifecycle through Platform API: mock order, payment callback, Gateway lease, settlement, usage review, refund review, and reputation update.

5. Usage review flow with review hash, six-dimensional metrics, and reputation signals.

6. Public deployment for judges to inspect the demo.

在 ETH Beijing 2026 期间，我们完成了 AgentLens 的 MVP（最小可行性产品）集成，核心功能如下：

1. Agent 目录与推荐系统： 建立了完善的 Agent 索引与智能推荐工作流。

2. 同类 Agent 对比页： 实现了基于运行安全边界标签的同类 Agent 对比功能。

3. 演示用 Agent 市场： 上线了不同状态的演示 Agent，包括：已识别平台镜像、未识别镜像以及外部工具集成状态。

4. 本地租赁全生命周期管理： 基于 Platform API 实现了完整闭环，包括：模拟订单、支付回调、网关租约、结算、租赁后评价、退款审核及信誉更新。

5. 租赁评价闭环： 完成了包含评价哈希、六维质量指标及信誉信号的评价处理流。

6. 公网部署： 已完成线上公网部署，供评委及用户现场查看与测试。

Not funded yet. Hackathon MVP / prototype stage.