Burnstack vs BurnLens
BurnLens, from Sairin Technology, is an open-source tracker for what AI agents and LLM calls cost: it reads Claude Code, Cursor and Codex logs, and its optional local proxy can cap each key's daily spend. Burnstack reads what all your dev vendors billed, AI included. Both keep your data local by default; they answer different questions.
The short version
- Choose BurnLens if you want cost per repository, developer or merged PR for coding agents; you need a hard daily cap per API key in front of OpenAI, Anthropic or Gemini; you'd rather run an open-source CLI and proxy you can read and self-host.
- Choose Burnstack if AI is one of several bills, next to hosting, databases, cloud, GitHub and Stripe fees; you want what each vendor actually billed, including usage that never went through a proxy; you want a desktop app with no code or base-URL changes.
At a glance
| Burnstack | BurnLens | |
|---|---|---|
| What it is | Desktop app (Mac; Windows soon) that reads your vendors' billing APIs | Open-source CLI and local proxy for AI agent and LLM costs (source) |
| Where your data lives | Your computer: keys in its keychain, history in a file you own | A SQLite file on your machine; optional cloud sync of cost metadata on paid plans (source) |
| What it covers | 32 vendors: hosting, databases, cloud, AI and dev tools, plus invoice imports | AI only: 10 LLM providers and Claude Code, Cursor, Codex and Gemini CLI logs (source) |
| How it gets costs | Vendors' billing and usage APIs; a few are estimated from usage and marked est. | Prices proxied calls and local agent logs itself |
| Can block spending | — | Yes: a daily dollar cap per key; the proxy answers 429 at 100% |
| Cost per merged PR | — | Yes, through the GitHub CLI |
| Cost per product | Applications across vendors | Tags: repository, team, customer, feature |
| Alerts | Spikes per vendor and model, idle resources, free minutes, overruns | Budget alerts at 50% and 80%, anomaly and waste detection |
| Setup | Paste a key per vendor; no code changes | pip install; point your SDK's base URL at the proxy, or just run the log scan |
| Open source | — | Yes, Apache 2.0 (source) |
| Price | Free forever; Enterprise for teams | Free; Cloud $29/month; Teams $99/month; Enterprise custom |
Who owns the data
BurnLens keeps what it records in a local SQLite database (~/.burnlens/burnlens.db) and says prompt bodies are never uploaded. Cloud sync, on the paid plans, sends token counts, costs, model names, opted-in tags and a prompt fingerprint, not prompt or response content (docs, FAQ). Your provider keys pass through its proxy on your machine if you use it. That's a genuinely local-first design too.
With Burnstack, API keys stay in your computer's keychain and your spend history is a file on your disk. You can export or delete all of it any time, and none of it is sent to us. What owning your data means.
Which bills each one reads
| Vendor | Burnstack | BurnLens |
|---|---|---|
| Vercel | ✓ | — |
| Neon | ✓ | — |
| Supabase | ✓ | — |
| OpenAI | ✓ | Proxied calls and agent logs, priced by BurnLens; not your OpenAI bill |
| Anthropic | ✓ | Proxied calls and Claude Code logs, priced by BurnLens; not your Anthropic bill |
| GitHub billing | ✓ | Merged PRs through the gh CLI, to price outcomes; not GitHub billing |
| OpenRouter | ✓ | — |
| DigitalOcean | ✓ | — |
| Twilio | ✓ | — |
| Stripe fees | ✓ | — |
| Cloudflare | ✓ | — |
| Netlify | ✓ | — |
| AWS | ✓ | Bedrock as a proxied provider (beta), not AWS billing |
| Datadog | ✓ | — |
| Sentry | ✓ | — |
| Cursor | ✓ | Local Cursor logs, priced by BurnLens; not your Cursor bill |
| PlanetScale | ✓ | — |
| Azure | ✓ | Azure OpenAI as a proxied provider (beta), not Azure billing |
| Google Cloud | ✓ | Gemini through the proxy, not Google Cloud billing |
| MongoDB Atlas | ✓ | — |
| Railway | ✓ | — |
| xAI | ✓ | xAI as a proxied provider (beta), not your xAI bill |
| Together AI | ✓ | Together as a proxied provider (beta), not your Together bill |
BurnLens lists ten LLM providers: OpenAI, Anthropic and Google stable; Groq, Together, Mistral, xAI, DeepSeek, Azure OpenAI and AWS Bedrock in beta (FAQ). We found no hosting, database, cloud-bill or GitHub-billing integration on its site, and no mention of reading providers' billing APIs or invoices: its FAQ says a call shows $0 when the model isn't in its pricing JSON.
Burnstack also reads Render, Heroku, Akamai (Linode), Vultr, Fastly, Runpod, fal.ai, Runway and Upstash (new in 0.9.0). We haven't checked BurnLens's coverage of those yet, so they aren't in the table.
Pricing
Free is $0 with the local proxy and dashboard, 1 API key, 1 seat, 7 days of history and 10,000 records a month. Cloud is $29/month (90-day history, up to 3 seats, a 7-day trial with a card), Teams $99/month (up to 10 seats, roles, 365-day history, audit log), and Enterprise is custom; annual billing saves about 28% (BurnLens pricing).
Burnstack is free forever. Enterprise (several accounts per vendor, monthly reports, team features) is priced per team. Pricing.
Where BurnLens is better
- Hard daily caps per API key: the proxy refuses calls once a key's recorded spend reaches its limit.
- Cost per repository, developer and merged PR for Claude Code, Cursor, Codex and Gemini CLI, read from logs you already have.
- Open source under Apache 2.0, so you can read and run all of it yourself.
- Ten LLM providers through one proxy, including Groq, Mistral, xAI and DeepSeek.
Where Burnstack is better
- The whole bill: Vercel, Neon, Supabase, AWS, Azure, Google Cloud, GitHub, Stripe fees and the rest of 32 vendors, not only AI.
- What you were actually billed, read from each vendor's billing API, so usage that never touched a proxy or an agent log still shows up.
- Invoice imports for tools with no billing API, and cost per product across vendors with applications.
- Alerts for idle Neon branches, paused Supabase projects and GitHub Actions free minutes.
- A desktop app for Mac: no Python, no proxy, no base-URL changes.
Not verified: BurnLens's site doesn't list supported operating systems (it needs Python 3.10 or newer) and doesn't say whether its dashboard runs locally or only in the cloud plans. Its cost-per-PR figure ($6.84 per accepted PR) is its own measurement on its own repository (GitHub).
They can work side by side: BurnLens to cap and attribute what your coding agents spend, Burnstack for the bill across every vendor.