AI Usage Dashboard

Real-time breakdown of all AI sources — local models, subscriptions & API calls

Last updated: April 7, 2026 · auto-refreshed by scheduled tasks
393
Dispatch Sessions
304
Last 7 Days
~62M
Est. Tokens (60d)
~1M
Tokens / Day
~$0
Monthly API Cost
3
Local Models
Interactive Charts

Token Distribution by Source

Daily Sessions — Last 7 Days

Cost Breakdown: Free vs Paid

Source Activity Over Time

Subscription Token Burn
~1M
Tokens / Day (current)
Post-cron migration rate
~7M
Tokens / Week
~200 sessions × varied load
~30M
Tokens / Month
Projected at current rate
~1.5M
Saved / Day (cron)
10-15 scheduled sessions eliminated
Note: These are subscription tokens, not API-billed. But they count against your daily message limits. Each Opus turn consumes ~2-4K input + ~2-4K output tokens. Heavy code sessions (worktree-based) can burn 500K-1M tokens each. The cron migration freed up ~1-1.5M tokens/day by moving repetitive tasks (market checks, briefings) off-platform.

Session Breakdown by Type (60 days)

Cron Migration Impact

● Before Cron Migration

Daily scheduled sessions10-15
Tokens/day (scheduled)~1.5M
Total tokens/day~2.5M
Message limit pressureHigh
Tasks platformAll Cowork

● After Cron Migration

Daily scheduled sessions~2-3
Tokens/day (scheduled)~100-150K
Total tokens/day~1M
Message limit pressureLow
Tasks platformCron + local scripts
Migrating scheduled tasks (market checks, N1 briefings, content pipelines) to system cron with local Gemma 4 and shell scripts reduced daily subscription token consumption by ~60%. Remaining Cowork sessions are interactive work and heavy code tasks that benefit from Opus-tier reasoning.
Cost Breakdown
SourceTypeModelEst. Tokens (60d)Monthly CostStatus
Cowork Tasks (~200 sessions)SubscriptionClaude Opus 4.6~30M$0 (Pro plan)active
Code Sessions (~50 sessions)SubscriptionClaude Opus 4.6~25M$0 (Pro plan)active
Scheduled Tasks (~143 sessions)SubscriptionClaude Opus 4.6~7M$0 (sandbox)migrating
Number One (Telegram)LocalGemma 4 E4B~2M$0 (llama.cpp)active
Newszor SummariesLocalGemma 4 E4B~500K$0 (was Gemini)active
Anthropic APIAPIClaude (fallback)~500K~$2minimal
Gemini APIAPIGemini 2.5 Flash~100K$0 (switched off)inactive
OpenAI APIAPIGPT-4o0$0 (no billing)inactive
Total~62M~$2/mo
AI Providers
🟣

Dispatch / Cowork

Claude Opus 4.6 · Subscription · ~62M tokens / 60 days
Primary AI workhorse. 393 total sessions, 304 in last 7 days. Estimated ~1M tokens/day across code, task, and scheduled sessions. These are subscription tokens — not API-billed, but they count against daily message limits.
subscriptionactive$0 API~62M tokens
💻

Code Sessions

Claude Opus 4.6 · Worktree-based · ~25M tokens est.
~50 heavy coding sessions (Oscar quiz, Newszor, OpenClaw, etc). Avg ~500K tokens each. Runs in isolated git worktrees as part of the Pro subscription.
subscriptionactive~25M tokens
🤖

Number One (Telegram)

Gemma 4 E4B · llama.cpp :8081
Personal Telegram bot running local Gemma 4 via llama.cpp. ~35 tok/s generation, ~103 tok/s prompt processing, 8192 context, 99 GPU layers.
localactive$0
📰

Newszor Summaries

Gemma 4 E4B · was Gemini API
News article summarizer. Recently switched from Gemini API ($) to local Gemma 4 model, eliminating API costs completely.
localactive$0 now
⏰

Scheduled Tasks

Claude Opus 4.6 · Cowork sandbox · ~7M tokens est.
~143 sessions × ~50K avg tokens. Mostly migrated to cron — saved ~1-1.5M tokens/day by eliminating 10-15 automated sessions daily. Remaining tasks run within subscription.
subscriptionmigrating to cron~7M tokens
🔑

Anthropic API

sk-ant-api03-t93h… · Claude fallback
Used only by OpenClaw as Claude fallback. 59 gateway log mentions. Minimal usage since Number One switched to local Gemma.
apiminimal
💎

Gemini API

Gemini 2.5 Flash · Now unused
44 total API calls from Newszor before switch to local Gemma 4. Service has been fully replaced — zero active usage.
apiinactive
🟢

OpenAI API

GPT-4o · Key set, needs billing
API key configured but billing not activated. 0 actual API calls made. Available as future fallback if needed.
apiinactive
Local Models (All Free)

Gemma 4 E4B PRIMARY

Runtime: llama.cpp
Port: 8081
Context: 8192 tokens
GPU Layers: 99
Prompt: ~103 tok/s
Generation: ~35 tok/s
Used by: Number One + Newszor

Ollama Models

Port: 11434
qwen3.5: available · fallback
gpt-oss:20b: available · fallback
qwen2.5:7b: available · fallback
Role: Backup / OpenClaw routing

LM Studio

Port: 1234
Model: GLM-4.7 Flash
Role: Available as fallback
Migration Timeline
Feb 2026
Claude Pro subscription started
Primary development tool. Dispatch/Cowork sessions begin accumulating.
Feb 2026
OpenClaw local gateway deployed
Mac Mini M4 running Ollama models. qwen3.5, gpt-oss:20b for daily automation.
Mar 2026
Gemini API used for Newszor
44 total API calls for news summarization. Gemini 2.5 Flash model.
Mar 2026
Number One Telegram bot launched
Personal AI assistant running on local Gemma model via llama.cpp.
Apr 2026
Newszor switched to local Gemma 4
Eliminated Gemini API costs. All summarization now runs on local hardware.
Apr 2026
Anthropic API usage declining
Only 59 gateway log mentions remaining. OpenClaw Claude fallback rarely triggered.
Apr 2026
Cron migration — scheduled tasks move off-platform
10-15 daily scheduled sessions (market checks, briefings, content pipelines) migrated to system cron + local scripts. Freed ~1-1.5M subscription tokens/day.
Apr 7, 2026
Current state: ~62M tokens over 60 days, 95%+ free
393 Dispatch sessions, ~1M tokens/day post-migration. Subscription-only for Opus tasks, local Gemma 4 for everything else.
Detailed Provider Pages
🟣

Claude Usage Details

Session history, project timeline, conversation replay
→
⚙️

OpenClaw Details

Gateway config, pipeline jobs, model routing
→
💎

Gemini Usage

API call history, migration to local
→
🔵

Codex Usage

Code generation stats, head-to-head comparisons
→
1n2.org · Dashboarder · Bitcoin Trains · Data Labs
Built with Claude Opus 4.6 · Powered by Gemma 4 + Ollama locally · April 2026