👁️ llmstat

See what your LLM calls actually cost — in real time.

A modern, playful terminal dashboard that sits in front of any OpenAI-compatible API (OpenAI, DeepSeek, OpenRouter, LiteLLM, vLLM, Ollama, your own gateway…) and shows you live token usage, cost, and latency. No SDK changes. No cloud. No account.

pip install llmstat → point your base URL at it → watch the numbers move.

Light

Dark

_{Two pastel themes — toggle with d. Today / all-time with t.}

Why

You're an indie dev shipping with LLMs — self-hosted models, a personal gateway, a couple of pay-as-you-go API keys — and you have no idea what you're spending until the invoice lands. The big observability platforms (Helicone, Langfuse…) are built for enterprise: hosted, signup-walled, dashboard-heavy, and hungry for your data. llmstat is the opposite — for the dev who just wants the number: one tiny local proxy, one SQLite file, one good-looking TUI.

• 🪶 Zero code changes — it's a drop-in proxy. Change one base URL.
• 🔌 Provider-agnostic — anything that speaks the OpenAI API works.
• 🌊 Streaming-aware — reads usage from SSE chunks, not just plain responses.
• 💸 Cost, not just tokens — built-in pricing table, override your own rates.
• 🔒 Local-first — your prompts never touch our anything. There is no "our anything".

Install

pip install llmstat

60-second tour (no API key needed)

llmstat demo      # seed some realistic sample traffic
llmstat dash      # open the dashboard
llmstat top       # or just a one-line status bar

Press t to flip between today and all-time, r to refresh, q to quit.

Real usage

1. Start the proxy in front of your provider:

llmstat serve --upstream https://api.openai.com
# llmstat proxy -> https://api.openai.com
# point your client base_url at: http://127.0.0.1:8787/v1

2. Point your client at it — only the base URL changes:

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8787/v1")  # was: the real URL
client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "hi"}],
)

3. Watch it live in another terminal:

llmstat dash

That's it. Every call your app makes now shows up with tokens, cost, and latency.

Tag your traffic (optional)

Send an X-Llmstat-Project header and llmstat will group spend by project — handy when one machine drives several apps.

Fix or add pricing

Models change prices constantly. Drop a ~/.llmstat/pricing.json to override:

{ "my-self-hosted-llama": [0.0, 0.0], "gpt-5.5-pro": [2.50, 15.00] }

Values are USD per 1M tokens: [input, output].

How it works

your app ──▶ llmstat proxy ──▶ real LLM API
                  │
                  └─▶ SQLite (~/.llmstat/usage.db) ──▶ TUI dashboard

The proxy forwards requests untouched and reads the standard usage block out of the response (for streams, it tees the SSE and grabs the final usage chunk). Nothing is buffered to disk except the counts you see.

Roadmap

• llmstat top — a one-line live status bar (--watch to refresh)
• Budgets & alerts (warn at $X/day)
• Export to CSV / Prometheus
• Per-project & per-key breakdown views

Ideas and PRs welcome — see an issue you'd like? Open one.

License

MIT © llmstat contributors