A Python-based CLI tool to aggregate LLM execution logs, distinguish internal/external users, and compute detailed multi-dimensional costs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install llm-usage-aggregator
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install llm-usage-aggregator using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is LLM Usage Aggregator?
The LLM Usage Aggregator is a robust utility designed to consolidate scattered LLM call logs from raw CSV files into structured, multi-dimensional Excel reports. By mapping model identifiers against a unified configuration schema (pricing_config.json), it accurately computes execution costs across diverse modalities including text, image, and video models.
This utility is a vital component of the Openclaw Skills ecosystem, offering teams transparent visibility into token consumption, image counts, and video generation times. It enables precise accounting of internal development versus external client usage to optimize API spend.
LLM Usage Aggregator Use Cases
- Aggregating messy raw CSV logs containing LLM invocation streams into clean, report-ready formats.
- Distinguishing between internal developer usage and client/external customer queries based on email domains.
- Tracking and auditing specific LLM performance and consumption metrics like prompt tokens, completion tokens, image generations, and video processing duration.
- Computing complex pricing rules for text tokens, static images, and variable-length audio/video outputs using unified model aliases.
How LLM Usage Aggregator Works
- Log Parsing: The tool accepts a CSV path and loads the usage logs, instantly resolving user identities using emails (falling back to phone numbers when emails are missing).
- User Segmentation: It scans email patterns to classify users as internal (matching domains like footprint, maybe, or fastest) or external.
- Pricing Calibration: It references
pricing_config.jsonto map raw model names to simplified aliases and identify their pricing type (per-million tokens, per-image, or per-second video rates). - Data Validation: Running pre-checks ensures no unconfigured models, invalid pricing schemas, or missing media dimensions slip through.
- Aggregation & Calculation: The engine aggregates totals and calculates exact USD costs across four major dimensions: user groups, platforms, specific models, and individual users.
- Excel Export: The system writes a single-sheet Excel report with four vertically stacked tables, each complete with automated totals.
LLM Usage Aggregator Setup
Ensure you have Python 3.x and pandas/openpyxl installed. Execute the command below using Openclaw Skills environment pathways:
python scripts/aggregate_llm_usage.py <csv_path> [output_path] [pricing_config_path]
LLM Usage Aggregator Data Schema & Taxonomy
Expected CSV Input Fields
| Field Name | Status | Description |
|---|---|---|
email or phone |
Required | Unique user identifier |
llm_provider |
Required | Hosting provider of the model |
llm_model |
Required | Raw model identifier (e.g., z-ai/glm-5v-turbo) |
prompt_tokens |
Optional | Number of input tokens |
completion_tokens |
Optional | Number of output tokens |
generated_image_count |
Optional | Number of images generated |
duration_seconds |
Optional | Duration of video generation in seconds |
generation_type |
Optional | Class of output (image, video, or empty) |
generate_audio |
Optional | Boolean flag for video audio generation |
Output Sheet Structure
The resulting Excel sheet contains four distinct tables structured sequentially:
- By User Type: Totals segmenting Internal vs. External usage.
- By Provider: Aggregations for each
llm_provider. - By Model: Usage broken down per
llm_modelalias. - By User: Detailed consumption per individual
email/phoneindex.
LLM Usage Aggregator Advanced Features
- Multi-Modal Pricing Models: Handles complex video pricing, auto-detecting audio streams to differentiate higher-priced with-audio generation versus basic video rendering.
- Model Aliasing: Automatically normalizes diverse model names (e.g., OpenRouter, Tokenhot, or Fal.ai endpoints) to streamlined, unified keys for aggregate accuracy.
- Comprehensive Pre-Flight Validation: Instantly logs anomalies such as unconfigured models, pricing mismatches, and text execution rows calculating to 0 USD cost to the console before finalizing output.
- Extensible Configuration: Seamlessly integrates with shared configurations within the Openclaw Skills suite, supporting newly added cutting-edge models like Kling v3, GLM 5V Turbo, and Doubao.
SKILL.md
Loading
Related Openclaw Skills
WeChat Search
A high-performance real-time search tool for fetching articles, videos, and media from the WeChat ecosystem without requiring a personal WeChat account.
InferSports Odds & Scores
A read-only, keyless CLI-driven AI agent skill providing real-time football and basketball scores, sharp betting lines, and de-vigged fair probabilities.
OpenClaw Traffic Guardian
OpenClaw Traffic Guardian is a baseline runtime security monitoring tool that provides opt-in HTTP/HTTPS proxy inspection, egress tracking, and policy review for AI agents.
Open Crawl
A high-performance web crawling and scraping skill designed for AI agents to retrieve structured Amazon product, search, and general public web page data instantly.
Alibaba Cloud AnalyticDB PostgreSQL Query Skill
A highly secure, read-only SQL query and data analysis capability for shell-capable AI agents connecting to AnalyticDB PostgreSQL databases via psql.
JisuAPI Agent
The JisuAPI Agent skill enables AI assistants to search and execute over 500 rich data services using simple natural language query matching.
Featured*
