AI Coding Assistant Analytics — One command to see exactly how much code AI wrote for you

15-Tool Unified · Line-Level AI Attribution · 600+ Model Cost Estimation · One-Click Weekly Report

What problem does it solve?

"How much code did AI write?" "Are these AI subscriptions worth it?" — Stop calculating manually. One command does it all.

Scenario	Solved by LumenCode
Precise AI Contribution	Not vague “AI helped a lot” — report AI-added and AI-deleted lines separately; contribution uses total changed lines (added + deleted). Every line accounted for.
Proving AI ROI	Auto-generated weekly report: “This week AI assisted 12 commits, added 3,180 lines, deleted 420 lines, cost $18.50.” Every number is traceable.
Weekly reports in 3 seconds	Pick period → click "Work Summary → Copy" → paste into Lark/DingTalk. Done in 3 seconds.
Per-project reporting	Configure multiple projects, then select one to generate an independent report for each project lead
Sprint cycle alignment	Beyond daily/weekly/monthly — pick any start/end date, no longer limited to fixed periods
Tracking AI costs	Built-in 600+ model pricing (incl. GLM, Kimi, Qwen, DeepSeek), auto-calculates equivalent API cost

How does it compare to `ccusage`?

Both LumenCode and ccusage read the same local logs from the same 15 agent CLIs. The difference is what you can do with the data — LumenCode adds a Web UI, an MCP server, and line-level AI attribution on top.

	ccusage	LumenCode
Interface	CLI	CLI + Web UI + MCP
Line-level AI attribution	—	✅ "this line was written by AI"
Publishable report	terminal / JSON	Markdown / Lark / DingTalk · Detailed / Brief
AI-generated smart report	—	✅ calls local agent for insights
Drill-down dashboard	—	✅ click chart → session / commit
Tools supported	15	15 (same set)
Cost & pricing	✅ offline + custom overrides	✅ 600+ models bundled (GLM / Kimi / Qwen / DeepSeek)

ccusage is a great, fast CLI — we draw inspiration from it. LumenCode reads the same ~/.claude logs, so both run side by side with no conflict.

Requirements

Node.js >= 20.0.0
Native SQLite dependency: better-sqlite3 (installed by npm install)
pm install)
At least one of the supported tools installed with existing session logs

Supported Tools & Data Directories

15 AI coding tools, all ✅ Fully Supported (session-level / token-level / model-level statistics).

Tool	Default Log Directory	Env Var (Optional)
Claude Code	`~/.claude`	`CLAUDE_DIR`
OpenAI Codex	`~/.codex`	`CODEX_DIR`
OpenCode	`~/.opencode`	`OPENCODE_DIR`
Gemini CLI	`~/.gemini`	`GEMINI_DIR`
Qwen Code	`~/.qwen`	`QWEN_DIR`
Goose	`~/.local/share/goose`	`GOOSE_DIR`
Amp	`~/.local/share/amp`	`AMP_DIR`
Hermes Agent	`~/.hermes`	`HERMES_DIR`
OpenClaw	`~/.openclaw`	`OPENCLAW_DIR`
Kimi CLI	`~/.kimi`	`KIMI_DIR`
Codebuff	`~/.config/manicode`	`CODEBUFF_DIR`
Droid	`~/.factory/sessions`	`DROID_DIR`
Pi Agent	`~/.pi/agent/sessions`	`PI_AGENT_DIR`
Kilo	`~/.local/share/kilo`	`KILO_DATA_DIR`
GitHub Copilot CLI	`~/.copilot/otel`	`COPILOT_OTEL_FILE_EXPORTER_PATH` / `COPILOT_DATA_DIR`

Without env vars set, the default directory is auto-detected. Multiple accounts can be comma-separated.

Get Started in 3 Seconds

# Global install (pin to latest)
npm install -g lumencode@latest
lumencode serve            # Start Web server, auto-opens browser

# Verify version (ensure ≥ 1.4.0)
lumencode --version

# Or run without installing
npx lumencode@latest serve

⚠️ Stuck on an old version? Run npm cache clean --force && npm install -g lumencode@latest to flush the cache and reinstall.

Zero-config out of the box — First run auto-detects all 15 tools' log directories above and derives project paths from session metadata. No manual setup needed.

Highlights

Core: line-level attribution × fifteen-tool unified × precise cost × one-click reports — accounting for AI coding's ROI down to every line, every cent.

Highlight	Description
🎯 Line-Level AI Attribution	Not "AI helped with this commit" — "This line was written by AI." Hook-based step tracking, precise down to every line
🌐 Fifteen-Tool Unified	Claude Code / Codex / Copilot and 12 more — all data auto-aggregated, one-click switch, cross-tool comparison
📝 One-Click Publishable Report	Detailed / Brief reports in seconds; Markdown / Lark / DingTalk formats, copy-paste ready, each section with insights
🤖 AI Smart Report	Calls one of local Claude Code / Codex / OpenCode to produce AI analysis (highlights, insights, risks, recommendations) from your stats and work report; supports Default and leadership-oriented "Workhorse" styles
💰 Precise Cost Estimation	600+ model pricing library (incl. GLM/Kimi/Qwen/DeepSeek) + Portkey API fallback; unknown models counted at $0 — real numbers only, no guessing
📂 Per-Project Reports	Multiple projects in parallel, each gets an independent report (commits + AI interaction + hotspot files)
📅 Sprint Cycle Alignment	Beyond daily/weekly/monthly — custom start/end dates to fit your iteration rhythm
🔍 Drill-Down & Trend Insights	Click any chart to reach session/commit; peak days, active streaks, 5-category tool usage at a glance
📦 Zero-Config Out of the Box	Auto-detects tool directories and derives project paths on first run — install and go
🌙 Light / Dark Theme	Dark mode default, all charts auto-adapt

Screenshots

Data Analysis Overview

Switch tools from the left sidebar. Main area shows Token usage, cost, model distribution, and AI contribution attribution.


Summary + Token Trends	Project Distribution + Hourly Activity + Session List

AI Contribution & Commit Analysis

Multi-Tool Dimension

Switch to "All Tools" view for cross-tool aggregate data and comparative analysis.

Multi-Tool Dimension

Project Distribution & Sessions

Per-project Token, cost, and session count stats. Click to drill down into individual session details.

Project Distribution & Sessions

Scenario Analysis

Categorize by work type (coding / testing / debugging / docs / review / planning), with matched keyword examples.

Scenario Analysis

Work Report · One-Click Publishable Weekly Report

Natural-language paragraph reports covering Token / cost / AI contribution / project highlights / code output, each section with insight commentary.

Detailed — Full data + insights + numbered sections, ideal for weekly/monthly reports
Brief — 3-5 sentence core summary, ideal for daily reports or group chat
Smart Report — Calls one of the local Claude Code / Codex / OpenCode agents from the page to generate AI analysis with data summary, work highlights, key insights, risks, and recommendations
Style Selection — Choose Default style, or "Workhorse" for a leadership-reporting tone before generation
Persistence & Freshness Hints — Smart reports are saved by period, project, report level, and style; stale source data prompts regeneration
Multi-Platform Format — Standard Markdown / Lark / DingTalk, one-click toggle
Per-Project — Select a project from the right panel to generate a project-specific report


Detailed	Brief

Light / Dark Theme

All chart colors auto-adapt for comfortable long sessions.

Light Mode

Dark mode is the default theme — the screenshots above were taken in dark mode.

Settings

Configure data sources (15 tool directories), enabled tools, cost mode, step-tracking attribution, scenario keywords, and appearance — all from the sidebar Settings page, organized into cards.

Settings Page

CLI Usage

lumencode <command> [period] [date] [options]

Command	Description
`serve`	Start Web server (default port 4567)
`report`	Generate CLI report (default command)
`init`	Initialize config file
`mcp`	Start MCP Server for Claude Code / Cursor etc. (see MCP Server)

Period	Description
`daily`	Daily report (default)
`weekly`	Weekly report (auto-calculates week range)
`monthly`	Monthly report (auto-calculates month range)

Examples

# Web mode (recommended)
lumencode serve

# CLI daily report
lumencode report daily
lumencode report daily 2026-05-15

# Weekly / Monthly
lumencode report weekly
lumencode report monthly 2026-05-01

# Specific projects only
lumencode report daily --projects D:/fzwork,E:/play/idea

# One-click publishable work summary
lumencode report daily --work          # Detailed
lumencode report daily --work --brief  # Brief
lumencode report weekly --work

MCP Server

LumenCode ships with a built-in MCP Server that exposes its AI coding analytics as 7 tools, callable directly from Claude Code / Cursor / Windsurf and other MCP-compatible clients — query usage, generate weekly reports, and analyze code contribution right in the conversation, no need to switch to the Web UI.

Tools

Tool	Description
`usage_summary`	AI usage overview: token consumption, cost, session count, model distribution
`daily_report`	Generate a usage report for a given date (Markdown)
`work_report`	Work summary (weekly/monthly), supports normal / brief / boss styles
`session_list`	List AI coding sessions within a time range
`trend_analysis`	Usage trends: daily token, cost, and request volume
`ai_contribution`	AI code contribution for a repo: contribution rate, commit attribution, hotspot files
`cost_breakdown`	Cost breakdown: per-model / per-project spend and cache hit rate

Configuration

Option 1: After global install (recommended)

npm install -g lumencode@latest

Add to your client's MCP config (Claude Code settings.json shown):

{
  "mcpServers": {
    "lumencode": {
      "command": "lumencode-mcp"
    }
  }
}

Option 2: Source / dev mode

{
  "mcpServers": {
    "lumencode": {
      "command": "node",
      "args": ["src/mcp/server.js"]
    }
  }
}

Cursor / Windsurf and other clients use the same mcpServers field — enter it via their respective settings. You can also run npm run mcp or lumencode-mcp directly in the foreground for debugging.

Highlights

Zero-config — Auto-detects all supported tools' log directories and derives project paths from sessions
stdio transport — Standard MCP stdio protocol; scans and caches logs on first call, reuses thereafter
Consistent results — All tools share the same lib/ stats and attribution implementations as the Web UI and CLI

Once configured, ask your AI assistant directly, e.g. "How much did AI coding cost me this week?", "Analyze AI contribution for the idea repo", or "Generate this week's work summary".

Configuration

First run auto-detects installed tools' log directories and project paths. For customization, open the Settings page from the left sidebar rail. Settings are organized into cards: Data Sources, Repositories, Cost & Billing, Attribution & Tracking, Scenario Keywords, and Appearance.

Setting	Description
Each tool's log directory	Data directories for the 15 tools, auto-detected by default per the table above; overridable in Settings or `config.json`
Enabled Tools	Specify which tools to enable, defaults to all detected
Local Project Paths	Git repo paths for code commit stats and AI attribution
Excluded Projects	Project names to exclude
Scenario Keywords	Work type classification keyword JSON
Cost Mode	Cost calculation source: `auto` (prefer log cost, fall back to pricing) · `calculate` (always recompute from token pricing) · `display` (raw log values only)
Step Tracking	Toggle step-level recording for line attribution (see Line-Level AI Attribution)
AI Attribution Params	Expert thresholds/weights for attribution scoring — read-only preview in UI; edit `config.json` directly to change

Line-Level AI Attribution (Optional)

Line-level attribution uses AI coding tool hooks to record file-edit steps, refining AI contribution from commit/file level down to line level. Claude Code uses PostToolBatch, Codex uses PostToolUse, OpenCode uses a project-level plugin. The feature is opt-in: without an initialized database, the hook silently skips and normal usage is unaffected.

# Run in the Git project root you want to track
node index.js hooks status
node index.js hooks enable       # Interactive tool selection, steps init, auto config backup

Enabling only modifies the current project's local config (.claude/settings.local.json, .codex/config.toml, .opencode/plugins/lumencode-step-tracker.js) — global config and other projects are untouched. To disable:

node index.js hooks disable

Data is written to .lumencode/steps.db in the current project. Existing .ccusage/steps.db files from older versions are copied to the new path on first use and kept as rollback-safe legacy backups.

Model Pricing Data

Local table — 590 models pre-synced from Portkey-AI/models with vendor canonical names
Alias mapping — 28 authoritative overrides mapping aggregator aliases (glm-5.1, kimi-for-coding) to correct pricing
API fallback — Unknown models auto-queried via Portkey's free API, results cached to data/pricing-cache.json; local + fallback covers 600+ models
Graceful degradation — When API is unavailable, the model is counted at $0 (won't be guessed), other models unaffected

FAQ

Issue	Solution
Browser shows "No Data"	First run will guide you through config; if skipped, open the Settings page (left sidebar)
Log directory not found on Windows	Default path is `C:\Users\<username>\.claude`, ensure `projects/` subdirectory exists
Port 4567 in use	Set env variable: `set LUMENCODE_PORT=8080 && lumencode serve`
Git stats not found	Project path is auto-derived from session `cwd`; if still unrecognized, set it manually in Settings
Cost showing $0	Model not in pricing table — try with network connection to let API fallback resolve, or add an `aliasOf` entry in `data/pricing.json` overrides
Smart report unavailable	Smart reports require one of local Claude Code / Codex / OpenCode — ensure the corresponding command is in your PATH

What's New

📖 Full changelog → Releases

Support This Project

If this tool helps you:

Star this repo — Help others discover it
File an issue — Report bugs or request features
Open a PR — Contributions welcome for model pricing, scenario keywords, or new tool adapters

lumencode

What problem does it solve?

How does it compare to `ccusage`?

Requirements

Supported Tools & Data Directories

Get Started in 3 Seconds

Highlights

Screenshots

Data Analysis Overview

Multi-Tool Dimension

Project Distribution & Sessions

Scenario Analysis

Work Report · One-Click Publishable Weekly Report

Light / Dark Theme

Settings

CLI Usage

Examples

MCP Server

Tools

Configuration

Highlights

Configuration

Line-Level AI Attribution (Optional)

Model Pricing Data

FAQ

What's New

Support This Project

License

Reviews (0)

What problem does it solve?

How does it compare to ccusage?

Requirements

Supported Tools & Data Directories

Get Started in 3 Seconds

Highlights

Screenshots

Data Analysis Overview

Multi-Tool Dimension

Project Distribution & Sessions

Scenario Analysis

Work Report · One-Click Publishable Weekly Report

Light / Dark Theme

Settings

CLI Usage

Examples

MCP Server

Tools

Configuration

Highlights

Configuration

Line-Level AI Attribution (Optional)

Model Pricing Data

FAQ

What's New

Support This Project

License

Reviews (0)

How does it compare to `ccusage`?