← AI Usage Dashboard
⚙️ OpenClaw Usage — Real Data Analytics
OpenClaw Gateway v2026.3.2 · Multi-model local + cloud · Feb 2 – Mar 8, 2026
5.77M
Total Tokens (verified)
Model Usage (from session files — real data)
Claude Sonnet 4.5
131 calls
Source: /Users/curiobot/.openclaw/agents/main/sessions/*.jsonl — 568 total model calls
Models Installed (Ollama — 100% Local)
| Model | Params | Size | Context | Calls | Role |
|---|
| qwen3.5 PRIMARY | 9B | 6.6 GB | 131K | 53 | Daily pipeline, summaries |
| gpt-oss:20b | 20.9B | 13 GB | 32K | 227 | Heavy reasoning, newspapers, reviews |
| llama3 | 8B | 4.7 GB | 8K | 1 | Fallback |
| nomic-embed-text | 137M | 274 MB | — | — | Embeddings |
Cloud Models Routed via OpenClaw
| Model | Provider | Calls | Role |
|---|
| Claude Sonnet 4.5 | Anthropic | 131 | Routed through gateway pre-billing-switch |
| Claude Opus 4.5 | Anthropic | 116 | Routed through gateway pre-billing-switch |
| GPT-5.2 | OpenAI | 15 | Fallback chain |
| GPT-4o-mini | OpenAI | 1 | Test |
LMStudio Model
| Model | Context | Max Output | API | Cost |
|---|
| GLM-4.7 Flash (zai-org) | 32,768 | 8,192 | OpenAI-compat (localhost:1234) | $0 |
Gateway Configuration
| Setting | Value |
|---|
| Version | OpenClaw v2026.3.2 |
| Port | localhost:18789 (loopback) |
| Auth | Token-based |
| Primary Model | ollama/qwen3.5 |
| Fallback | ollama/gpt-oss:20b |
| Channels | Telegram (DM pairing, partial streaming) |
| Max Concurrent | 4 agents, 8 subagents |
| Hardware | Mac Mini M4 — Metal GPU |
⚠️ Known Issue: Brave Search API Key Missing
web_search returns missing_brave_api_key. Fix: run openclaw configure --section web and provide a Brave Search API key, or set BRAVE_API_KEY in Gateway environment.
Daily Pipeline Jobs (15 total via master-run.sh)
| # | Job | AI Model | Status |
|---|
| 1 | Data Labs Collect (YouTube, Curio, articles) | — | ✅ |
| 2 | City News (5 cities) | — | ✅ |
| 3 | Entertainment (movies, theater, museums) | — | ✅ |
| 4 | Ego Search (Thomas Hunt, Curio Cards) | — | ✅ |
| 5 | Tweetster (996 accounts) | — | ✅ |
| 6 | Facebooker | — | ✅ |
| 7 | Curio Charts + Archive exports | — | ✅ |
| 8 | CurioQuant signals | — | ✅ |
| 9 | Article Archive + indexer | — | ✅ |
| 10 | Dashboarder generate + deploy | — | ✅ |
| 11 | Morning Summary → Telegram | qwen3.5 | 🤖 AI |
| 12 | Daily THunt newspaper | gpt-oss:20b | 🤖 AI |
| 13 | Daily Logs | — | ✅ |
| 14 | Curio Review (Mondays) | gpt-oss:20b | 🤖 AI |
| 15 | Deploy to Droplet (SCP) | — | ✅ |
Pipeline Log History
| Date | Log Size | Notes |
|---|
| Feb 2 | 1.0 MB | First major session (earliest session file) |
| Feb 9 | 523 KB | MediaLog project session |
| Feb 18 | 18 KB + 9.6 KB | First automated pipeline run + daily briefing |
| Feb 20 | 11 KB | Full newspaper generated (1,700 words) |
| Feb 21–28 | ~700 B/day | Lightweight Curio Oracle runs (npm missing) |
| Mar 1–4 | ~700 B/day | Same — oracle builds failing |
| Mar 5 | 36 KB | Pipeline expansion — new scrapers added |
| Mar 6 | 90 KB | Full 15-job run (608 minutes!) |
| Mar 7 | 18 KB | Optimized run — all 15 jobs |
Session Files (Real Token Data)
| Session | Size | Date | Primary Model |
|---|
| e00bcdb9… | 1.0 MB | Feb 2 | Mixed (Claude + Ollama) |
| e91b83c3… | 1.0 MB | Feb 9–24 | gpt-oss:20b (newspapers) |
| a71b7699… | 523 KB | Feb 9 | Claude Sonnet 4.5 |
| 7aad73f0… | 185 KB | Mar 4 | qwen3.5 |
| ef505555… | 84 KB | Mar 1 | gpt-oss:20b |
| cd86d05f… | 54 KB | Feb 25 | qwen3.5 |
| 3cd90f7b… | 52 KB | Mar 6 | qwen3.5 |
| 7b41f1c8… | 27 KB | Mar 8 | qwen3.5 (Brave error) |
Cost Analysis
| Category | Tokens | Cost |
|---|
| Local inference (Ollama: qwen3.5 + gpt-oss:20b + llama3) | ~2.8M | $0 |
| Cloud via gateway (Claude Sonnet/Opus, pre-switch) | ~2.7M | Included in Anthropic plan |
| OpenAI fallback (GPT-5.2) | ~0.27M | Included in OpenAI plan |
| Total verified | 5.77M | ~$0 marginal cost |
💰 If all 5.77M tokens were via Claude API ($3/$15 per M tokens): estimated ~$50+. Running locally + existing plans = $0 additional.