Make the map better

Your machine
is evidence.

Every community benchmark begins inside the ToshLLM app. The app runs the shared workload, lets you inspect the exact sanitized data, signs it with your local benchmark identity, and submits it only after you confirm.

Download ToshLLM
01

Run in the app

ToshLLM records the run.

Open Benchmarks → Share with the community in ToshLLM, choose any local model, and stop the local server before measuring so both processes do not compete for VRAM.

  • Any local or split GGUF model
  • Standard pp512 and tg128 workload
  • Three independent repetitions
  • Your real GPU and runtime configuration
02

Review

See every byte first.

Before anything is uploaded, ToshLLM displays the exact JSON it will sign and send: model artifacts, hardware, configuration, measurements, engine hash and sanitized raw output.

No browser upload. No hidden collection.Local paths, account names, prompts and chat content are excluded. Creating a benchmark identity and contacting the service happen only after your explicit share action.
03

Sign & submit

The app brings the receipts.

When you confirm, ToshLLM signs the reviewed payload with a private key stored in your Keychain and sends it through the benchmark API. No GitHub account is required. Consistent results are published automatically as app-recorded community data; extreme or irregular measurements wait for manual review. Authorized ToshLLM Lab runs can be published as verified.

Get the app

One trusted path

From your Mac.
Through the app.

01RunInside ToshLLM
02ReviewExact sanitized JSON
03SubmitSigned community result