Glazyr Viz
AI 与智能体by glazyr
Glazyr Viz 直接将 Chromium 原始内存帧提供给 AI agent，绕过脆弱的 DOM scraping 与 Cloudflare 封锁，实现 177 FPS 零拷贝视觉与原生 USDC 结算。
什么是 Glazyr Viz？
Glazyr Viz 直接将 Chromium 原始内存帧提供给 AI agent，绕过脆弱的 DOM scraping 与 Cloudflare 封锁，实现 177 FPS 零拷贝视觉与原生 USDC 结算。
核心功能 (10 个工具)
get_optic_nerve_statusReturns a high-level dashboard of the agent's visual health, including FPS, latency, and Aquarium population metrics.
browser_navigateDispatches a navigation command to the agent's browser, used to switch between benchmarks or sites.
browser_set_fish_countControls the hardware load by setting the number of active WebGL fish in the Aquarium simulation.
peek_vision_bufferPerforms a low-latency 'peek' at the raw vision stream, returning resolution, sequence numbers, and optional base64 frame data.
browser_evaluate_jsEvaluates arbitrary JavaScript in the GCP Big Iron browser context.
run_dogfood_surgeExecutes the standardized dogfooding sequence: sets baseline, triggers a 30k fish surge, and returns status.
verify_paymentVerifies a USDC transfer on the Base network to grant vision credits to the current session (1 USDC = 1,000,000 frames).
get_remaining_creditsRetrieve the current balance of cognitive frames available for this session.
browser_clickLegacy control: Dispatches a mouse click to the specified coordinates.
browser_typeLegacy control: Types the specified text into the active browser element.
README
Glazyr Viz: 7.35ms Perception. 90%+ Token Savings. 🚀
Ditch the screenshot loop. Glazyr Viz is a high-performance Chromium fork that provides agents with Zero-Copy Vision—direct, raw memory access to the frame buffer for sub-10ms perception.
🎯 Real-World Use Cases
- High-Density Data Extraction: Navigating complex tables, Canvas-based charts, and WebGL interfaces where DOM scrapers fail.
- Latency-Critical Automation: Executing multi-step workflows (checkout bots, form filling) at human or super-human speeds.
- Large-Scale Scraping: Reducing API tokens by 99%, allowing for thousands of perception cycles at a fraction of the cost.
- Anti-Bot Resilience: Interacting with raw coordinates to bypass detection systems that flag standard WebDriver behavior.
⚡ Performance Floor
- 7.35ms Latency: Sub-10ms frame-to-data conversion floor.
- 99% Token Savings: 12-16 tokens per perception cycle via the
vision.jsonschema. - Zero-Jitter: Synchronous frame access directly from the Chromium Viz subsystem.
Installation
# Copy the skill to your OpenClaw skills directory
cp -r glazyr-viz ~/.openclaw/workspace/skills/glazyr-viz
# Install dependencies
cd ~/.openclaw/workspace/skills/glazyr-viz/scripts
npm install
Quick Start
# Navigate to a page
node skills/glazyr-viz/scripts/navigate.js https://news.ycombinator.com
# Extract data (Ah-Ha Demo)
node skills/glazyr-viz/scripts/showcase.js
Pricing (Launch Tiers)
| Tier | Frames | Price |
|---|---|---|
| Free | 2,500 | $0 |
| Developer | 100,000 | $3 |
| Professional | 500,000 | $15 |
Get your API key at glazyr.com/dashboard.
📘 Technical FAQ: Zero-Copy Vision
Q: How do you achieve 99% token savings?
Most agents use "Pixel-Pushing"—they capture a screenshot, encode it to Base64, and send the entire image to an LLM. This consumes roughly 1,200–1,600 tokens per frame. Glazyr Viz uses the vision.json schema to extract semantic UI metadata and raw coordinate vectors directly from the Chromium Viz subsystem’s frame buffer. This reduces the payload to 12–16 tokens per perception cycle.
Q: Is any "intelligence" lost by not sending a full screenshot?
None. In fact, you gain precision. Standard vision models often "guess" coordinates from pixels, leading to click hallucinations. vision.json provides the exact [x, y] coordinates and semantic metadata (ARIA roles, labels, states) directly from the Chromium render tree. Your agent doesn't have to guess; it knows.
Q: How does this eliminate "Jitter"?
Traditional "screenshot" methods are asynchronous. Because Glazyr Viz is baked into the Chromium source, frame access is synchronous. The agent perceives the UI state at the exact moment the frame is committed to the GPU. Build fast. Stop serializing. Built by MAGNETAR SENTIENT L.L.C. // V1.0.0 General Release
常见问题
Glazyr Viz 是什么？
Glazyr Viz 直接将 Chromium 原始内存帧提供给 AI agent，绕过脆弱的 DOM scraping 与 Cloudflare 封锁，实现 177 FPS 零拷贝视觉与原生 USDC 结算。
Glazyr Viz 提供哪些工具？
提供 10 个工具，包括 get_optic_nerve_status、browser_navigate、browser_set_fish_count 等。
相关 Skills
Claude接口
by anthropics
面向接入 Claude API、Anthropic SDK 或 Agent SDK 的开发场景，自动识别项目语言并给出对应示例与默认配置，快速搭建 LLM 应用。
✎ 想把Claude能力接进应用或智能体，用claude-api上手快、兼容Anthropic与Agent SDK，集成路径清晰又省心
提示工程专家
by alirezarezvani
覆盖Prompt优化、Few-shot设计、结构化输出、RAG评测与Agent工作流编排，适合分析token成本、评估LLM输出质量，并搭建可落地的AI智能体系统。
✎ 把提示优化、LLM评测到RAG与智能体设计串成一套方法，适合想系统提升AI开发效率的人。
智能体流程设计
by alirezarezvani
面向生产级多 Agent 编排，梳理顺序、并行、分层、事件驱动、共识五种工作流设计，覆盖 handoff、状态管理、容错重试、上下文预算与成本优化，适合搭建复杂 AI 协作系统。
✎ 帮你把多智能体流程设计、编排和自动化统一起来，复杂工作流也能更稳地落地，适合追求强控制力的团队。
相关 MCP Server
顺序思维
编辑精选by Anthropic
Sequential Thinking 是让 AI 通过动态思维链解决复杂问题的参考服务器。
✎ 这个服务器展示了如何让 Claude 像人类一样逐步推理，适合开发者学习 MCP 的思维链实现。但注意它只是个参考示例，别指望直接用在生产环境里。
知识图谱记忆
编辑精选by Anthropic
Memory 是一个基于本地知识图谱的持久化记忆系统，让 AI 记住长期上下文。
✎ 帮 AI 和智能体补上“记不住”的短板，用本地知识图谱沉淀长期上下文，连续对话更聪明，数据也更可控。
PraisonAI
编辑精选by mervinpraison
PraisonAI 是一个支持自反思和多 LLM 的低代码 AI 智能体框架。
✎ 如果你需要快速搭建一个能 24/7 运行的 AI 智能体团队来处理复杂任务（比如自动研究或代码生成），PraisonAI 的低代码设计和多平台集成（如 Telegram）让它上手极快。但作为非官方项目，它的生态成熟度可能不如 LangChain 等主流框架，适合愿意尝鲜的开发者。