CostCanary — LLM Cost Linter for VS Code

Detects LLM calls in your code, shows inline cost annotations, and offers one-click fixes to reduce spend.

What it does

• Inline cost annotations above every LLM call: 🐦 🟢 ~$0.003/call · 847 calls/30d · "chat"
• Squiggly-line diagnostics for untracked calls, missing userId, exceeded cost thresholds, and placeholder values
• Quick Fixes (Cmd+.) to wrap calls with track(), add userId, or switch to a cheaper model
• Model autocomplete with pricing as you type inside model: "..." or model="..."
• Save-time warnings when a file has untracked expensive calls (configurable: off/warn/block)
• Status bar showing the per-file cost estimate for the active document
• Sidebar with three panels: Features (cost by feature), Savings (model downgrade recommendations), Alerts

How it works

CostCanary scans your JavaScript, TypeScript, and Python files for LLM API calls using regex pattern matching. It annotates each call with cost data from your CostCanary account (via /api/features). When a cheaper model is available or a call is missing cost tracking, it surfaces a diagnostic and a Quick Fix — no manual configuration required beyond signing in.

Requirements

• VS Code 1.85.0 or later
• Node.js 18+ (for JS/TS projects)
• A CostCanary account — sign up at costcanary.dev

Quick start

1. Install the extension from the VS Code Marketplace
2. Open the Command Palette (Cmd+Shift+P) and run CostCanary: Sign In
3. Paste your API key from costcanary.dev/settings
4. Open any JS/TS/Python file that makes LLM API calls — cost annotations appear automatically

Detected patterns

JavaScript / TypeScript

• track({ feature: "...", userId, fn }) — CostCanary SDK
• wrapClient(client, { ... }) — CostCanary SDK
• client.chat.completions.create(...) — OpenAI
• client.responses.create(...) — OpenAI
• client.embeddings.create(...) — OpenAI
• client.images.generate(...) — OpenAI
• client.messages.create(...) — Anthropic
• model.generateContent(...) — Google Generative AI

Python

• client.chat.completions.create(model="...") — OpenAI
• client.messages.create(model="...") — Anthropic
• model.generate_content(...) — Google Generative AI
• ChatOpenAI(...), ChatAnthropic(...) — LangChain
• OpenAI(model="...") — LlamaIndex
• client.chat(model="...") — Mistral

Diagnostics

All diagnostics except placeholderUserId can be dismissed per-file via Dismiss for this file in the Quick Fix menu, or reset globally with CostCanary: Reset All Dismissals from the Command Palette.

Configuration

Keyboard shortcuts

Quick Fixes are standard VS Code code actions — they appear via the lightbulb icon or the keyboard shortcut when the cursor is on a diagnostic.

Commands (Command Palette)

Sidebar panels

Open the CostCanary activity bar icon to see three panels:

• Features — Tracked features ranked by total cost, with cost-per-call, call volume, and % of bill
• Savings — Model downgrade recommendations from the API, with estimated monthly savings. Session fixes are shown in the header.
• Alerts — Configured alert rules and recent trigger history

Troubleshooting

Annotations don't appear

• Confirm you are signed in: run CostCanary: Sign In from the Command Palette
• Check that costcanary.showCodeLens is true in settings
• Annotations only appear on files with detected LLM calls

"No cost data yet" for a feature

• Data appears after events are ingested by the SDK. Call track() in production and wait for the next ingest cycle.

The model switch Quick Fix replaced the wrong text

• The fix only targets model: "..." or model="..." property values — not variable names or comments. If it replaced something unexpected, file a bug at github.com/costcanary/costcanary/issues.

Dismissed diagnostics won't come back

• Run CostCanary: Reset All Dismissals from the Command Palette to restore all diagnostics.

Extension shows stale data

• Run CostCanary: Refresh Data to force a re-fetch from the API.

Links

• Marketplace page
• Documentation
• Dashboard
• Report an issue