Dashboard walkthrough

The dashboard lives at tensorloop.tech/dashboard. It has four pages: Overview, Keys, Analytics, and Activity.

Overview

The landing page after sign-in. Four hero tiles plus a recent-activity strip.

Hero tiles

• 30-day spend — USD across all your keys for the rolling 30-day window. Numbers update within seconds of a call landing.
• Requests (30d) — count of inference requests in the same window.
• Active keys — number of non-revoked, non-expired keys you currently hold.
• Top model — the model you used most by request count, with a small sparkline of its daily call counts.

The window is rolling 30 days, not calendar month. Spend on day 31 drops off automatically.

Recent calls

Bottom strip on the Overview page — the most recent 8 calls. Use it to confirm a new integration is landing. For more depth, jump to Activity.

Keys

Where you mint, view, and revoke API keys. Two sections:

Create key

Right-side panel. Fields:

• Alias — a short, dashed name (my-laptop, prod-backend). Lowercase letters, numbers, dashes only. Must be unique within your account.
• Models — multi-select of models your plan allows. On free, only gpt-4o-mini. On pro, the full upstream catalog.
• Budget (USD / 30d) — must be ≤ your plan's per-key cap.
• RPM — requests per minute cap; must be ≤ your plan's per-key cap.
• Expiry — when the key stops working. Default 90 days, max 365.

After clicking Create API key, the plaintext appears in a reveal dialog once. Copy it to a password manager immediately — it is not retrievable.

Keys table

Left side. One row per active key. Columns:

• Alias — what you named it.
• Models — the allowlist set at mint time.
• Budget — used / cap with a fill bar. Red badge when exhausted.
• RPM — the rate limit set at mint time.
• Spend (30d) — rolling USD against this key.
• Expires — date the key will stop working.
• Actions — trash icon to revoke.

Click revoke, type the alias to confirm, and the key is immediately and permanently disabled. Applications using it start getting 401 within seconds.

Analytics

Spend and request volume over time.

• Daily spend (30d) — bar chart, one bar per day. Hover for exact USD.
• Daily requests (30d) — companion bar chart.
• Peak day — the highest-spend day in the window, called out separately.
• By model — pie split of which models you're spending on.

Use this to spot drift (a new integration suddenly accounting for 90% of spend) or to confirm a quiet day (no calls = no chart).

Activity

Per-request log of the most recent 20 calls. Use it to debug specific requests.

Each row has:

• Timestamp — when the call landed.
• Key alias — which key it was billed to.
• Model — which upstream model served the request.
• Tokens — prompt / completion / total split.
• Cost — USD, to four decimal places.
• Status — green for success, red for upstream errors. Failed calls still appear here but don't count against the budget.

This is the page you want open when integrating a new application — every successful call shows up within a few seconds.

Plan and account

Plan tier is shown in the sidebar banner ("Free plan — 2 keys, gpt-4o-mini"). Sign-out is in the bottom-left corner.

Currently the only plan change is free → pro and that's a manual upgrade — there's no self-serve flow yet.

What's next

• Authentication — full key lifecycle.
• Errors — what every error code means.
• Troubleshooting — common integration issues.