Skip to content

Tellama v1.3.2

Choose a tag to compare

@azossy azossy released this 22 Jul 13:38

Tellama v1.3.2 focuses on the on-device chat experience: faster first response, longer-lasting context, and smoother model handling.

Highlights

  • Faster first token (TTFT). The system prompt is now frozen for the duration of a chat session, so the prompt prefix stays byte-stable across turns and the native KV cache prefix is reused instead of re-processed. Previously, automatic memory changed the system prompt every turn, which forced a full re-prefill and made the first token slower as the conversation grew. Newly captured memories apply from the next conversation.
  • Context lasts longer. Because the memory set no longer grows mid-session, it stops eating into the conversation history budget, so earlier turns stay in context longer.
  • Keyboard no longer covers the input. On Android 15 edge-to-edge, the chat input now rises above the on-screen keyboard (explicit enableEdgeToEdge() plus IME-aware layout), across Android 10–15.
  • Your last model comes back. On relaunch, Tellama warm-loads the model you were last serving in the background, so it is ready when you open a chat instead of starting unloaded.
  • Clearer model errors. A model that fails to load now shows its message in a dialog (with the error code) instead of a banner at the top of the list that scrolls out of view.

APK integrity

  • File: tellama-v1-3-2.apk
  • SHA-256: 43fd6c4739a199b5680dda50b30fe371370144cd1609eb3777d2a621369f46d2
  • Package: com.tellama.app
  • Version: 1.3.2 (versionCode 32)
  • Signer certificate SHA-256: 5c3180b0290135491a0dec1dba6acab9119a8e17a60c8663d390bb2073007516

Updating

Tellama v1.2.1 and later check the public release feed and validate the APK SHA-256, package, higher version, and signing certificate before opening Android's installer.