MODEL FINDER / BY INDIGLOO
Understand your
model footprint.
Explore usage exports and estimate model-weight memory. Your file stays in this browser tab; no provider connection or API key is needed.
USAGE ANALYTICS
Your activity, in view.
No account is connected. Import your own JSON or load the synthetic example.
Maximum 2 MiB and 10,000 records. Download format example · View import format
Activity by date (UTC)
Model breakdown
| Model | Requests | Input tokens | Output tokens | Reported USD | Cost coverage |
|---|
MODEL-SIZE ANALYSIS
Estimate the memory
behind the parameters.
This is a model-weight estimate, with an overhead allowance you choose. It is not a hardware benchmark or a guarantee that a model will run. A handset with about 3 GB of RAM is the screen. Several department assistants and your business files stay on a server you control or a private cloud.
Weight GiB = parameters in billions × 1,000,000,000 × bits ÷ 8 ÷ 1,073,741,824. The allowance multiplies that estimate by (1 + overhead ÷ 100). KV cache, context length, concurrency, quantization metadata, activations, and runtime allocations may need additional memory. MoE models require total stored parameters, not only active parameters.
YOUR DATA, YOUR CONTROL
A small, explicit import format.
Use a JSON array with one object per request. Required: a timezone-qualified timestamp, session ID, model name, and non-negative integer token counts. USD cost is optional; omit it or use null if unknown. Extra fields are ignored and not retained in dashboard records. Remove prompts, responses, secrets, and identifying details before importing.
[{
"timestamp": "2026-10-12T10:00:00Z",
"session_id": "session-1",
"model": "Your model name",
"input_tokens": 1000,
"output_tokens": 300,
"cost_usd": 0.002
}]No automatic access to Claude, ChatGPT, or provider usage is included. Imported data stays in memory until you clear it, reload, or close the page. If you share analytics with a host assistant, that is a separate action governed by the host’s policies.