Cofferwise · LLM FinOps for AI startups

Wise choices for your AI spend.

Per-feature, per-user, per-prompt cost attribution across OpenAI and Anthropic. Slack alerts on anomalies. Five-minute setup.

Install the CLI →Open dashboard

Available today · free · open source

Audit your AI bill from Claude Code.

coffer-cost-review is a Claude Code skill. It reads your code with semantic understanding — knows when cache_control is applied at runtime, when an agent loop is bounded externally, when an f-string is actually static at init — and only flags real cost waste.

# install
npm install -g coffer-cost-review

# then open Claude Code and ask:
# "review my LLM costs"

No Node? One-line shell install: curl -fsSL https://cofferwise.com/install-skill.sh | sh

npm →GitHub →

The SDK integration

For production tracking, one-line wrapper.

Drop in coffer.openai.OpenAI instead of openai.OpenAI. Tag each call with feature and user. We never sit in your critical path — uploads run in a background thread, silent on failure.

from coffer.openai import OpenAI

client = OpenAI(
  api_key="sk-...",
  coffer_api_key="cof_...",
  default_metadata={"feature": "chatbot"},
)

response = client.chat.completions.create(
  model="gpt-4o-mini",
  messages=[{"role": "user", "content": "hi"}],
  _coffer={"user_id": user.id},
)

Across providers

OpenAI today. Anthropic and Bedrock next. One dashboard for every model your app touches.

Attribution, not just totals

Per feature, per user, per prompt template. Spot the free user burning your budget.

Advisory, not autonomous

We tell you which prompts to switch to Haiku. We never re-route your traffic without you.