编程

TinkerClaw Command Center

试用

Stop guessing what your AI costs. Tinker shows every token, every dollar, every context byte — in real time.

它能做什么

**Your $200 Opus session didn't have to happen.** Tinker shows you exactly where every token goes — before the bill arrives.

技能文档

Tinker Command Center

Your $200 Opus session didn't have to happen. Tinker shows you exactly where every token goes — before the bill arrives.

The Problem

You switched to Claude API because Anthropic did the right thing. Respect. But now you're running Opus through OpenClaw and a single deep conversation burns $20+ in tokens with zero warning. You check your dashboard three days later and wonder what happened.

That's not a billing problem. That's a visibility problem.

What Tinker Does

Tinker is a real-time command center that sits on top of your OpenClaw gateway. It shows you what's filling your context window, what each response costs, and where your budget stands — live, as it happens.

🗺️ Context Treemap

Interactive squarified treemap of your context window. See exactly what takes space: system prompt sections, conversation history, tool results. Drill down from categories → messages → raw text. When you wonder "why is my context 180K tokens?" — this tells you in one glance.

📊 Response Treemap

Same visualization for model output. How much is text, how much is thinking, how much is tool calls? Per LLM call within a run, so you see the real cost of that 8-step tool loop.

💰 Live Cost Tracking

Per-provider token usage. Daily and monthly estimates. The 5-hour Claude rate-limit window with countdown timer. You'll never hit a rate limit by surprise again.

⚠️ Budget Alerts

Set a monthly limit. Get warned at 70%, 90%, and 100%. No more "I'll check later" — later is too late with Opus.

🔄 Multi-Call Run View

When your agent loops through tools (search → read → edit → test → commit), each call's context and cost is broken out individually. You'll see which tool calls are token hogs and which are cheap.

💬 Full Chat Interface

Not just a dashboard — it's a complete webchat with session switching, tool call inspection (expand any tool inline), and real-time streaming. Use it as your daily driver or just for monitoring.

Pricing Reference

These are the API costs Tinker tracks:

ModelInput (per 1M)Output (per 1M)Watch out?
Claude Opus 4 / 4.5$15.00$75.00⚠️ Yes. One deep session = $20+
Claude Sonnet 4 / 3.5$3.00$15.00Sweet spot
Claude Haiku 3.5$0.80$4.00Background tasks
Gemini 3 Pro$1.25$5.00Great failover
Gemini 2 Flash$0.10$0.40Near-free

Install

Tinker is a bundled OpenClaw plugin in the globalcaos fork. It ships as part of the build.

# Clone the fork
git clone https://github.com/globalcaos/tinkerclaw.git openclaw
cd openclaw

# Build (includes Tinker UI)
pnpm install
pnpm build

# Access
# Production: http://localhost:18789/tinker/
# Development: cd tinker-ui && pnpm dev → http://localhost:18790/tinker/

Architecture

tinker-ui/                ← Standalone Vite + Lit app (zero upstream conflicts)
├── src/
│   ├── app.ts            ← Main shell: sidebar + panels + WebSocket client
│   └── panels/
│       ├── context-treemap.ts    ← What fills your context window
│       ├── response-treemap.ts   ← What each response costs
│       └── context-timeline.ts   ← Context usage over time
├── index.html
└── vite.config.ts

extensions/tinker/        ← OpenClaw plugin (serves UI from gateway)
├── index.ts
└── openclaw.plugin.json
  • Zero file overlap with upstream ui/ — no merge conflicts, ever.
  • ~3,300 lines of focused TypeScript + Lit components.
  • No external services — connects to your local gateway WebSocket on port 18789.

Why Now?

Anthropic refused the Pentagon contract. Respect. Claude overtook ChatGPT in the App Store this weekend. People are migrating to Claude API in droves.

But Claude API through OpenClaw is unmetered. There's no $20/month cap. Opus input costs $15 per million tokens, output costs $75. A single agent session with tools can easily consume 200K+ tokens. Do the math.

If you're switching to Claude API because you believe in Anthropic's mission — you should also know what you're spending. That's not about being cheap. That's about being informed.

Tinker makes you informed.

Get the fork


Built by globalcaos · Part of The Tinker Zone

相关技能

以 AI 机器人身份加入视频会议,提供语音、虚拟形象与屏幕共享四种模式。

作者 johnpatternai21 次安装8 星标

在本地磁盘以分类纯 Markdown 文件保存需要长期留存的事实,与智能体内置记忆并存。

作者 Iván1 次安装

把自然语言描述转为结构化 JSON,并由 mcp-diagram-generator MCP 服务生成 Draw.io、Mermaid 或 Excalidraw 图表文件。

作者 nssa.io1.0k 次安装47 星标

从 AdMapix API 拉取广告创意、应用、榜单和收入预估等数据,原样返回结构化 JSON。

作者 fly0pants

诊断生产力系统反复失效的根因,给出最小干预——容量测算、瓶颈定位、可靠的本地记录。

作者 Iván1 次安装

按用户明确指令,在得到大脑(Get笔记)中保存、搜索并管理笔记与知识库。

作者 iswalle763 次安装66 星标

globalcaos 的更多技能

浏览全部技能

为 OpenClaw 智能体加上 JARVIS 风格的英式嗓音与冷面幽默,一条命令同步输出语音和紫色聊天气泡。

作者 globalcaos190 次安装4 星标

OpenClaw 原生 WhatsApp 通道:22 项消息与群组操作,叠加 Protocol v2 多智能体协同。

作者 globalcaos116 次安装3 星标

一个本地面板,统一追踪 Anthropic、Gemini、OpenAI、Manus 四家提供商的 token 用量与预算告警。

作者 globalcaos59 次安装

Stop sending 'format this JSON' to Opus. Stop sending 'cron job' to GPT. Billing-aware routing guide for choosing among the models already configured in your OpenClaw setup when assigning an agent, sub-agent or cron task — flat-rate first, metered only when justified, budget pressure respected. Not for picking models outside your configuration, and not a runtime proxy.

作者 globalcaos16 次安装

Your agent says 'done' — but did it check? Superpowers turns any OpenClaw agent into a disciplined engineer. Verification iron law (evidence before claims), three-agent code review (build → verify spec → verify quality), systematic debugging (4-phase root cause, three-strike rule), brainstorming gates (design before code), and anti-over-engineering rules. Use when: (1) coding tasks of any complexity, (2) debugging failures, (3) about to claim work is complete, (4) spawning sub-agents, (5) planning features, (6) reviewing code. Inspired by top coding agent methodologies, adapted for OpenClaw multi-agent architecture.

作者 globalcaos14 次安装

Read and search Outlook, inspect attachments, and create or edit drafts without any send endpoint. Uses one short-lived Microsoft Graph access token supplied on stdin for one run; it never stores credentials. Bulk mailbox export is opt-in.

作者 globalcaos23 次安装