> ## Documentation Index
> Fetch the complete documentation index at: https://docs.toolken.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# How Toolken Works
> The request path, what you can do today, and what's coming.
Toolken sits between your app and your LLM provider. One base-URL change is all it takes. Every request is forwarded with your own provider key, and its cost, attribution, and latency are recorded along the way.
## The request path
```mermaid theme={null}
%%{init: {"theme":"base","themeVariables":{"primaryColor":"#0f172a","primaryTextColor":"#e2e8f0","primaryBorderColor":"#334155","lineColor":"#64748b","fontSize":"14px","fontFamily":"ui-sans-serif, system-ui"}}}%%
flowchart TD
A["Your app"] -->|"base_url + X-Toolken-Key + provider key"| G["Toolken gateway\n(edge)"]
G -->|"provider key forwarded untouched"| P["LLM provider"]
P -->|"response"| G
G -->|"response streamed back"| A
G -. "records cost & attribution" .-> D["Dashboard"]
```
Toolken adds a single edge hop. Recording happens off the response path, so it does not slow your call.
## Roll your own vs Toolken
| | Roll your own | Toolken |
| ----------------- | ------------------------------------------------------------------ | ------------------------------------- |
| Per-agent cost | Parse usage from every response, join to a pricing table, store it | One `X-Toolken-Metadata-Agent` header |
| New model pricing | Track every provider price change yourself | Priced at the gateway, kept current |
| Latency added | None (you are inline) | One edge hop, off the response path |
## What you can do today
Slice any metric by agent, feature, or customer.
Request-rate and concurrent-stream limits at the edge.
One endpoint, your keys.
Cost, usage, reliability, performance.
## Coming soon
Cut cost and latency on repeated calls.
Automatic retry across providers.
Issue keys with attribution and limits built in.
## Next
Make your first attributed request in under two minutes.