01
Follow the session, not the invoice.
Reconstruct working windows from request timing and metadata. See context growth, model changes, cache behavior, and errors turn by turn.
Token intelligence for LLM work
See where every
token goes.
TknScope sits between your tools and model providers, turning every request into clear, session-level usage intelligence—without changing the response your client receives.
Provider agnosticStreaming preservedContent off by default
TknScope
Production · 24 hours
Proxy live
USAGE OVERVIEW
Last 24h⌄Token activity
Total tokens2.84M↑ 18.4%
Requests1,28442 sessions
Avg latency412msp95 1.2s
Usage heatmapTokens by hour
MonTueWedThu
0006121824
Active sessionsNow
ANTHROPICOPENAISTREAMING SSESESSION WINDOWSUSAGE EXTRACTION
The product
From raw requests
to a legible system.
Usage data is only useful when it retains the shape of the work that produced it.
02
Find the hot spots.
A shared heatmap shows when and where token space lights up—without another dashboard for every provider.
03
Break usage down.
Group by provider, model, session, key, endpoint, or time. Compare the same signal from the angle your question needs.
How it works
One change.
A complete signal.
01
Route once
Point your existing client at the TknScope proxy.
02
Pass through
Auth, request bodies, and streamed responses stay intact.
03
Read the shape
Provider usage becomes sessions, heatmaps, and breakdowns.
YOUR CLIENTSDK / agent / app
request →TKNSCOPEtransparent proxyusage ↓
→ streamPROVIDERAnthropic / OpenAI
Privacy boundary
Count the tokens.
Leave the words alone.
Prompts, completions, and API keys never become logs or metric labels. TknScope records the usage facts the provider returns. Content capture is a separate, explicit gate—and it ships off.
privacy.conf
capture_content = offpersist_api_keys = neverlog_prompt_text = neverusage_counts = observed Privacy boundary active
TknScope