TokenStation5 01.004.001

Security fixes in 01.004.001

This update fixes security holes in TokenStation5's built-in Tailscale. Update if you use Tailscale to reach your PS5 (#97).

• Your tailnet's access rules now apply. Before, every device on your tailnet could reach the API, whatever the rules in your Tailscale admin console said. Now only devices your rules allow can connect, and none can until the rules arrive from Tailscale.
• A way around the API key is closed. A device on the tailnet that opened a connection and dropped it at the right moment could be let in without a key. Over Tailscale, every request now needs a key, as intended.
• HTTPS no longer trusts the leaked eMudhra certificates (Tailscale's bulletin TS-2026-013), for every connection the title makes: Tailscale, downloads and search.
• Your Tailscale sign-in key is deleted from the settings file once the PS5 has signed in, so it can't be read later over the home network's FTP.

About "Tailworm": those flaws are in Tailscale SSH and Serve in Tailscale's own tailscaled. TokenStation5 runs its own small Tailscale client, without SSH or Serve, so they don't apply to it.

Tested on a PS5 over Tailscale: a valid key works for chat, and no key, a wrong key or a deleted key is refused.

Also in this version: everything new since 01.003.000

Google's Gemma 3 12B

• Gemma 3 runs on the PS5 (#94). Gemma 3 12B (bartowski's google_gemma-3-12b-it-Q4_K_M.gguf, 6.8 GB) passed every GPU check against the CPU and 59 of 60 app-style checks, tool calls and a write-a-file agent task included. Find it in the Models page's Hugging Face search.
• Most of its layers only look back 1,024 tokens, so they keep a small rolling memory, and it fits about 29,700 tokens of context next to its weights.
• It reads prompts at 250 tokens a second and writes about 40 a second, up from 188 and 35 in its first build on the console.

Faster and steadier

• New GPU kernels for wide attention heads and sliding windows: prompts read a third faster on Gemma, and writing is 13% faster (#94).
• Switching models no longer fails with "out of GPU memory" after a bigger model was loaded: the model's memory is now split into smaller pieces (#94).
• A GPU check on the model you're serving no longer runs out of memory (#94).

Better answers out of the box

• Known-good settings (#92): each checked model (Qwen 3 8B, Qwen 2.5 Coder 7B, Llama 3.1 8B, Llama 3.2 3B, Mistral Nemo 12B and now Gemma 3 12B) gets the sampling settings that scored best on the PS5. An app's own settings still win.
• Four fixes to how tool calls and thinking are read back from the models (#92).
• Mistral Nemo 12B takes tools in its own format (#91).
• The Models page and Hugging Face search only offer models the engine can run (#90).

Your models survive updates

• The first start after a store update brings your downloaded models back, through the jailbreak's FTP server (#93). Start the title once after updating. Settings, keys and chats were already kept.

Install

Install or update from the Prospero Store, or as before: PPSA99555.zip holds the PPSA99555/ app folder.