Xula Herald हिन्दी

Agent

Local AI chat, your models, and practical Android tools

MobileFork

Screenshots

About this app

Agent is an independently maintained Android app for local AI chat and practical device workflows. Choose a model first: open Settings > Models to import a GGUF, LiteRT-LM (.litertlm), or Android .task file with the system file picker. Local import requires neither an account nor an internet connection. Choose an imported or downloaded model and use the start action; the app selects the matching runtime. Model compatibility and available memory still determine whether a model can run. Optional online providers and downloads are separate from local files. Response preferences, runtime controls, and advanced GGUF settings are grouped under clearly labelled expandable sections. Six interface languages, Android accessibility labels, and confirmation before removing a model help keep controls understandable. The Full edition includes Android tools for files, app launch, accessibility-assisted actions, notifications, shortcuts, widgets, and automations. Permissions and remote processing remain opt-in. The Play edition has a more restricted tool set. Agent is not an official Nous Research product and is not endorsed by Nous Research or Teknium. Open-source credits and licenses are retained. The existing application identifier is kept so installed users retain their upgrade path and app data. Save a diagnostic report without starting a model: Settings > General > Diagnostics and logs > Export diagnostic log. Android lets you choose the folder and filename. Reports include current memory and the last model-start attempt. Review the redacted file before sharing; nothing is uploaded automatically.

APK versions

APK files are served by the original repository.

0.13.161337.0 MiB Minimum SDK 24
Download APK
Release date · 2026-10-29Architecture · arm64-v8a, x86_64Target SDK · 36
Release notes

Recover local model startup when another process occupies the preferred port. Update LiteRT-LM to 0.18.1, stable llama.cpp to 0.7.0, and TurboQuant. Refresh security dependencies and strengthen context-file injection checks. Preserve advanced model arguments and safely clean up failed LiteRT initialization.

SHA-256 checksum45c83ba0579975a8b4aa834cb5d245463f1f389e325bc37b3810064fc9dfbb3c
0.13.160334.3 MiB Minimum SDK 24
Download APK
Release date · 2026-10-04Architecture · arm64-v8a, x86_64Target SDK · 36
Release notes

Sharper glass borders and smoother page transitions. Faster native startup by avoiding duplicate environment setup. Updated stable and experimental GGUF runtimes. Added context-budget and GGUF CPU-thread controls, plus clearer RAM-override diagnostics. Turbo3 is for experimental GGUF; context capacity depends on the model and available RAM.

SHA-256 checksuma27cab4f397fab329d71503c882372091beb3c128b7f4e0830dfe0fe618c9559
0.13.159333.3 MiB Minimum SDK 24
Download APK
Release date · 2026-10-03Architecture · arm64-v8a, x86_64Target SDK · 36
Release notes

Security maintenance: update the embedded HTTP client and transport; improve Android-only validation. Workspace import and memory-diagnostics fixes are included. Experimental Gemma KV-cache investigation is documented separately; no blanket quality or memory guarantee. Save a diagnostic log from Settings > General to a folder and filename you choose, even before a model starts. Includes device and memory details with known-secret redaction.

SHA-256 checksum0e6444687f7f2668c5ad15398130d0071490b182b496704d60c24349eedf5b0c