它将两个主要 AI 编码助手工具的代币、成本和生产力指标统一到一个自托管堆栈中，提供自动化的 Grafana 仪表盘和严格的隐私过滤。
Local metrics pipeline for Claude Code and OpenAI Codex CLI:
Claude Code ─┐
├─ OTLP ─> OpenTelemetry Collector ─> Prometheus ─> Grafana
Codex CLI ───┘
Runs entirely on your machine. Published ports bind to 127.0.0.1 by default.
Client authentication stays in Claude Code or Codex; this stack needs no
provider API key.
One Grafana dashboard, "Token & Usage Monitor", provisioned automatically. Five rows:
| row | what it answers |
|---|---|
| Overview | session starts, tokens, estimated cost, Codex threads, active time, cache hit rate |
| Token Usage | tokens over time, split by type, model, and Claude effort level |
| Cost Analysis | estimated cost over time and by model, across both vendors |
| Agents | tokens and estimated cost for Claude traffic carrying a named agent |
| Sessions & Productivity | commits, pull requests, lines changed, edit decisions |
Cost split by model across both vendors is the reason this exists rather than either vendor's own view. Claude figures are the provider's own estimates; Codex figures are list-price estimates computed here — see Cost meaning for what that difference implies.
Overview headline tiles:
Token usage over the selected range, split by token type:
The Codex side of the cost split, priced from the list-price rules:
Named-agent attribution — the share of Claude tokens carrying an agent name, and its split by agent type:
Every panel is empty until a client sends data. Setup steps 3 and 4 do that.
ai_* metric namespace.| Component | Supported requirement |
|---|---|
| Host | macOS or Ubuntu only |
| Containers | Docker with docker compose |
| Claude Code | 2.1.214 or newer for corrected streaming token/cost accounting |
| Codex CLI | 0.145.0 or newer for GPT-5.6 cache-write telemetry |
| Python | 3.10 or newer for hook installation, pricing maintenance, and tests |
| Git | Needed for cloning and the optional commit metric |
| Password g |
根据分类、Topic 和编程语言匹配的相似项目。
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.