inferwise

Smart model selection and cost enforcement for LLM API calls.

Inferwise scans your codebase for LLM API calls, recommends the cheapest model that can handle each task, estimates per-token costs, and enforces budget guardrails. Works with any CI system or locally as a git hook.

Note: Inferwise tracks pay-as-you-go API costs (billed per token to your API key). It does not track flat-rate subscriptions like Claude Code, Cursor, Copilot, or ChatGPT Plus.

Quick Start

# See what your LLM calls cost
npx inferwise estimate .

# Get smart model recommendations
npx inferwise audit .

# Set up guardrails: config + git hooks + CI
npx inferwise init

Or install globally:

npm install -g inferwise

Commands

inferwise init

Set up config file, git hooks (husky/lefthook/plain git), and print CI setup instructions for GitHub Actions, GitLab CI, Bitbucket, and more.

inferwise estimate [path]

Scan for LLM API calls and estimate costs.

inferwise estimate .
inferwise estimate ./src --volume 5000
inferwise estimate . --format json

inferwise diff [path]

Compare token costs between two git refs. Enforces budget policy from inferwise.config.json.

inferwise diff
inferwise diff --base main --head HEAD
inferwise diff --fail-on-increase 500

inferwise check [path]

Verify total LLM costs are within budget. Exits with code 1 if exceeded. For AI agents and automation.

inferwise check . --max-monthly-cost 10000
inferwise check . --max-cost-per-call 0.05

inferwise calibrate [path]

Fetch real usage from provider APIs (Anthropic, OpenAI, OpenRouter) and compute correction factors for more accurate estimates.

ANTHROPIC_ADMIN_API_KEY=sk-ant-admin-... inferwise calibrate .
OPENROUTER_API_KEY=sk-or-... inferwise calibrate .  # All providers via OpenRouter
inferwise calibrate . --dry-run

inferwise audit [path]

Find cost optimizations with smart, capability-aware model recommendations. Infers what each LLM call does from prompts in your code and suggests cheaper models that can handle the task — with reasoning and confidence levels.

inferwise fix [path]

Auto-apply model swap recommendations from audit. Rewrites model IDs in source files.

inferwise fix .                    # Apply all recommendations
inferwise fix . --dry-run          # Preview without modifying files
inferwise fix . --min-savings 500  # Only apply fixes saving >$500/mo

inferwise price [provider] [model]

Look up model pricing. Compare models side-by-side. Designed for humans and AI agents.

inferwise price anthropic claude-sonnet-4
inferwise price --compare anthropic/claude-sonnet-4 openai/gpt-4o
inferwise price --list-all

Budget Enforcement

Add budgets to inferwise.config.json (created by inferwise init):

{
  "budgets": {
    "warn": 2000,
    "block": 50000,
    "requireApproval": 10000,
    "approvers": ["platform-eng"]
  }
}

• warn — flags the PR with a warning label
• block — fails the CI check, blocks merge (emergency brake)
• requireApproval — requests review from approvers before merge

SDK (Programmatic API)

import { estimateAndCheck } from "inferwise/sdk";

const result = await estimateAndCheck("./src", { maxMonthlyCost: 10000 });
if (!result.ok) {
  console.error("Over budget:", result.violations);
}

Pure data, no console output, no process.exit — safe for embedding in agent orchestration, pipelines, or automation.

Supported Providers

Anthropic, OpenAI, Google AI, xAI, Perplexity — with LangChain, Vercel AI SDK, AWS Bedrock, Azure OpenAI, and LiteLLM pattern detection.

File types: .ts, .tsx, .js, .jsx, .mjs, .cjs, .py

Documentation

See the main repo for full documentation, CI setup guides, and estimation methodology.

License

Apache 2.0