cost-audit-agent

English | 中文

Solo Founder OS agent #11 — monthly bill audit across the SaaS stack indie founders use. Flags underused subscriptions, spend spikes, and specific waste recommendations with estimated dollar savings.

Built by Alex Ji. Born from this thought:

I'm subscribed to 14 SaaS tools. I have no idea which I actually use enough to justify. The 2nd of every month I should know exactly what's worth keeping.

What it does

Pulls month-to-date usage + spend from each provider, runs heuristics, outputs a markdown report with specific dollar-tagged recommendations:

## 🔍 Waste findings — potential savings: $42.00/mo

- ⚠️ **[vercel]** Vercel Pro underused (~$20.00/mo)
  - Only 3 deployment(s) this month on a $20.00/mo Pro plan. If you can wait 45s
    longer per build, the Hobby plan is free.

- 💡 **[anthropic]** Sonnet-heavy usage — consider Haiku for routine tasks (~$22.00/mo)
  - 73 Sonnet 4.6 calls ($33.40) vs only 12 Haiku calls. For diff review, summary,
    classification — Haiku is 3× cheaper with comparable quality.

Install

git clone https://github.com/alex-jb/cost-audit-agent.git
cd cost-audit-agent
pip install -e .
cp .env.example .env  # fill in VERCEL_TOKEN etc.

Usage

# One-shot report to stdout
cost-audit-agent

# Write to your Obsidian vault on the 2nd of each month (cron-friendly)
cost-audit-agent --out ~/Documents/Obsidian/.../cost-$(date +%Y-%m).md

# Limit to one provider
cost-audit-agent --provider vercel

Providers (v0.1)

Provider	What it reads	Auth
vercel	deployments + build-minute estimate + Pro-plan baseline	VERCEL_TOKEN
anthropic	local agent usage logs (~/.build-quality-agent/usage.jsonl, ~/.funnel-analytics-agent/usage.jsonl); org-wide if admin key provided	optional ANTHROPIC_ADMIN_KEY + ANTHROPIC_ORG_ID

Why local-logs fallback for Anthropic

Anthropic's organization billing API requires an admin-scoped key, which most indie founders don't bother creating. But build-quality-agent and funnel-analytics-agent both write per-call usage to ~/.<agent>/usage.jsonl. For a solo founder running 2-3 agents, those logs cover ~95% of monthly Anthropic spend. We aggregate them by month with the published Haiku/Sonnet/Opus prices.

Roadmap

• v0.1 — Vercel + Anthropic providers · markdown report · waste heuristics · 14 tests
• v0.3 — OpenPanel, HyperDX, Supabase, GitHub Actions providers
• v0.4 — MCP server: query monthly spend + top savings from Claude Desktop
• v0.5 — 3-month trailing baseline · spike detection
• v0.6 — Claude-summarized executive narrative at top of report
• v0.7 — Auto-cancel button (one-click webhook to provider's cancel endpoint when finding savings > $10/mo, gated by HITL like vc-outreach)

MCP server (Claude Desktop / Cursor / Zed)

Ask your AI assistant "what's my biggest waste this month?" and let it call the auditor.

pip install 'cost-audit-agent[mcp]'

Then in ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "cost-audit": {
      "command": "cost-audit-mcp",
      "env": {
        "VERCEL_TOKEN": "...",
        "VERCEL_PLAN_USD": "20",
        "ANTHROPIC_ADMIN_KEY": "...",
        "ANTHROPIC_ORG_ID": "...",
        "OPENPANEL_CLIENT_ID": "...",
        "OPENPANEL_CLIENT_SECRET": "...",
        "HYPERDX_API_KEY": "...",
        "SUPABASE_PERSONAL_ACCESS_TOKEN": "...",
        "SUPABASE_PROJECT_REF": "...",
        "GITHUB_TOKEN": "...",
        "GITHUB_REPO": "alex-jb/vibex"
      }
    }
  }
}

Tools:

• get_monthly_report() — full markdown audit, all providers
• get_provider(name) — one provider's spend + waste findings
• top_savings(n=5) — top N findings by estimated $/mo, sorted

License

MIT.

🧩 Part of the Solo Founder OS stack

A growing collection of MIT-licensed agents that share solo-founder-os as their base — cron, eval, reflexion, AnthropicClient, HITL queue, notifiers. Each agent is independently useful; together they cover the full one-person company workflow.

🌐 The whole stack is live in production at vibexforge.com.

Agent	What it does
solo-founder-os	The shared base lib (cron · eval · reflexion · skill library · DGM-lite). Every other agent depends on it.
build-quality-agent	Pre-push Claude diff reviewer + local build runner — catches CI-killing changes before they ship.
customer-discovery-agent	Reddit pain-point scraper + Claude clustering for product validation.
funnel-analytics-agent	Daily founder brief + real-time PH-day alerts across 9 sources.
orallexa-marketing-agent	AI marketing agent — submit project once, get platform-native posts for X / Reddit / HN / Dev.to / 小红书 + 7 more. Powers vibexforge.com.
vc-outreach-agent	Cold email drafter — investors (vc mode) or paying customers (customer mode, merged from customer-outreach in v0.9.0). HITL queue + SMTP sender.
bilingual-content-sync-agent	EN ⇄ 中文 i18n diff + Claude translate + HITL apply. Batch API path @ 50% off.
customer-support-agent	Triage user messages → auto-draft replies → HITL queue. Closes the L5 customer-support layer.
payments-agent	Overdue-invoice reminder drafter. Stripe-shaped types + MockProvider fallback + hard money-safety prompt rules.

Each agent's own row is omitted from its README. Install whichever solve real problems for you — pip install <agent-name>.