Why We Built a Daily Spend Dashboard for Token-Based AI Tools — The $3,600 Surprise That Changed Everything

TL;DR: June 2026: Copilot AI Credits ($0.04/1k tokens) + Anthropic API + OpenAI API = three token-based bills, zero visibility mid-month. Our $380/mo predictable spend jumped to $3,600+ projected. We built a daily cron dashboard aggregating claude-code usage, gh copilot usage, and OpenAI usage API — now alert at 80% budget. Dashboard code →

The Context

Two-dev team, 120 hrs/mo AI coding across 5 codebases. Pre-June 2026: Copilot Business ($38/mo flat) + Anthropic API direct ($150–300/mo). June 1: Copilot switches to AI Credits (~$0.04/1k tokens for agentic). June 15: Claude Code credit pool launches ($100/mo for ~100 credits/day). OpenAI API unchanged (per-token). Three different token economies, three different billing cycles, zero unified view. May bill: $380. June projected (first 2 weeks actuals): $1,800 → $3,600/mo annualized.

What We Tested

Tracking Approach	Tools Covered	Verdict	Why
Manual spreadsheet (weekly)	All three	❌	1-week lag; surprises still happen
Individual tool dashboards	Each separately	❌	Context-switching; no aggregate view; different units (credits vs tokens vs $)
Custom cron + Slack alerts (our build)	All three unified	✅	Daily 5pm rollup; single $ view; alerts at 80% threshold
Third-party (Langfuse, Helicone)	API-only (Anthropic/OpenAI)	❌	Misses Copilot credits; no Claude Code CLI integration; $50+/mo

The Pivot Point

June 8, 2026: Mid-month check. Copilot credits: 2.4M tokens used ($96) — on track for $192/mo vs $19 prior. Anthropic API: $420 (Opus-heavy debugging). OpenAI: $45 (Codex included in Plus). Total: $561 in 8 days → $2,100/mo projected. No single tool warned us. Copilot’s “300 credits included” = 7.5M tokens, but agentic sessions burn 500k–1M each. We’d exhausted included pool by June 12. Realization: Token billing is opaque by design. You need daily rollups in dollars, not tokens/credits, with a single budget threshold across all tools.

What We Use Now

Daily Spend Dashboard (.toolcrucible/token-spend-dashboard.sh, runs via cron 5pm daily):

#!/bin/bash
# Unified daily spend tracker for token-based AI tools
# Outputs: Slack alert if >80% daily budget; logs to ~/.toolcrucible/spend.log

DATE=$(date +%Y-%m-%d)
BUDGET_DAILY=120  # $120/day = ~$3,600/mo ceiling
THRESHOLD_PCT=80

# 1. Claude Code credit pool (via CLI)
CC_USAGE=$(claude-code usage --json --since yesterday 2>/dev/null || echo '{}')
CC_CREDITS=$(echo "$CC_USAGE" | jq -r '.credits_used // 0')
CC_USD=$(echo "$CC_USAGE" | jq -r '.usd_estimate // 0')

# 2. GitHub Copilot AI Credits (via gh API)
GH_USAGE=$(gh api /user/copilot/usage --jq '.[] | select(.date=="'"$DATE"'") | .tokens_used' 2>/dev/null || echo 0)
GH_USD=$(echo "scale=4; $GH_USAGE * 0.00004" | bc)  # $0.04/1k tokens

# 3. OpenAI API (via usage endpoint, requires org key)
# OAI_USAGE=$(curl -s -H "Authorization: Bearer $OPENAI_KEY" \
#   "https://api.openai.com/v1/usage?date=$DATE" | jq '.total_usd // 0')
OAI_USD=0  # Codex covered by ChatGPT Plus $20 flat

# Aggregate
TOTAL_USD=$(echo "scale=2; $CC_USD + $GH_USD + $OAI_USD" | bc)
THRESHOLD_USD=$(echo "scale=2; $BUDGET_DAILY * $THRESHOLD_PCT / 100" | bc)

# Alert
if (( $(echo "$TOTAL_USD > $THRESHOLD_USD" | bc -l) )); then
  slack-post "#dev-alerts" "💸 AI Spend Alert: $${TOTAL_USD}/$${BUDGET_DAILY} daily (${THRESHOLD_PCT}% threshold). CC: $${CC_USD} (${CC_CREDITS} credits), GH: $${GH_USD} (${GH_USAGE} tokens)"
fi

# Log
echo "[$DATE] Total: $${TOTAL_USD} | CC: $${CC_USD} (${CC_CREDITS} cr) | GH: $${GH_USD} (${GH_USAGE} tok) | OAI: $${OAI_USD}" >> ~/.toolcrucible/spend.log

Budget protocol:

• Daily ceiling: $120 (allows bursts, caps month at ~$3,600)
• Alert at 80% ($96): Triggers “Opus flag review” + “Copilot agentic pause”
• Weekly review (Mon 9am): cat ~/.toolcrucible/spend.log | tail -7 → adjust allocations
• Hard cap: If daily > $150, auto-disable Copilot agentic + force Opus flag for remainder of day

When You’d Choose Differently

• Single-tool teams: Individual tool dashboard suffices (Claude Code usage, Copilot gh api, OpenAI usage page).
• Enterprise with negotiated flat-rate: Token tracking irrelevant if fixed contract.
• Light usage (<$100/mo total): Manual weekly check fine; dashboard overhead not worth it.
• Teams using Langfuse/Helicone for observability: Extend existing pipeline rather than build custom.

Tool Crucible Rating

Dimension	Rating (1–5)	Notes
Overall	4	Solves real pain; DIY maintenance burden
Ease of Use	3	Requires CLI/API access for each tool; bc math fragile
Value	5	Prevented $2,000+/mo surprise; zero SaaS cost
Support	N/A	Internal tool; evolves with provider APIs

---

This is part of our AI Coding Tool Evaluation series. See full dashboard code: Token-Based AI Spend Dashboard 2026: Unified Daily Tracking Across Copilot, Claude, OpenAI