边缘计算AI

Build

1. Open in Android Studio (Ladybug+ recommended), or use Gradle CLI.
2. Build:

./gradlew :app:assembleDebug

1. Install:

./gradlew :app:installDebug

Curl Examples

ADB Port Forwarding

To test the API locally from your host machine over USB/WiFi:

adb forward tcp:8455 tcp:8455

#### Health Check

curl http://127.0.0.1:8455/health

#### List Available LLMs (OpenAI Format)

curl "http://127.0.0.1:8455/v1/models?type=llm"

#### Generate TTS WAV to File

curl -s -X POST "http://127.0.0.1:8455/v1/audio/speech" \
  -H "Content-Type: application/json" \
  -d '{"input":"Welcome to Nabu on device AI","engine":"kokoro","response_format":"wav"}' \
  --output test_speech.wav

#### Simple OpenAI Chat Completion

curl -X POST "http://127.0.0.1:8455/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"gemma3-1b-it-q4",
    "messages":[{"role":"user","content":"Name three fast animals."}],
    "stream":false
  }'

#### Stream OpenAI Chat Completion

curl -N -X POST "http://127.0.0.1:8455/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"gemma3-1b-it-q4",
    "messages":[{"role":"user","content":"Say hello in five words."}],
    "stream":true
  }'

#### Send OpenCode Tool Call Request

curl -X POST "http://127.0.0.1:8455/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma3-1b-it-q4",
    "messages": [
      { "role": "user", "content": "What is 55 times 12?" }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "multiply",
          "description": "Multiply two numbers",
          "parameters": {
            "type": "object",
            "properties": {
              "a": { "type": "number" },
              "b": { "type": "number" }
            }
          }
        }
      }
    ],
    "stream": false
  }'

Demo Video

• Watch the demo video

Screenshots

<p align="center"> <img src="https://github.com/user-attachments/assets/fa7e8816-41f4-48eb-83c3-29aac0f98251" alt="Mixer" loading="lazy" /> <img src="https://github.com/user-attachments/assets/882237cc-1c0f-40d7-bdb0-2a528846ae32" alt="Conversation Settings" loading="lazy" /> <img src="https://github.com/user-attachments/assets/a34eda22-09a9-4d29-9e26-9ab9208eb3eb" alt="Settings" loading="lazy" /> <img src="https://github.com/user-attachments/assets/da19d300-2a05-4f2b-9b49-0cc0547aa79f" alt="Basic Screen" loading="lazy" /> </p>

Local API Server

Nabu includes an opt-in local REST API server for on-device inference, exposing both text-to-speech and an OpenAI-compatible /v1/chat/completions endpoint.

• Default bind: 127.0.0.1:8455
• Optional LAN bind: 0.0.0.0:8455 (enable in Settings)
• Security note: there is no API auth layer yet; use LAN exposure only on trusted networks.

Enable it from Settings: - Enable API Server - Expose API on LAN (optional)

Playground Workflows

• TTS workbench: switch engines (kokoro, supertonic, soprano) and compare runtime behavior on-device.
• LLM workbench: run local chat models from managed LiteRT .task downloads or imported .gguf files.
• Book workflow: open documents, edit text, save projects/bookmarks, and pre-generate per-line WAV audio for offline playback.
• Chat + TTS loop: generate responses with local LLMs and speak them through the active TTS engine.

Audiobook Workflow File Types

Unknown/other file types fall back to plain text extraction.

Model Listing

Endpoint paths for checking loaded/downloaded resources: - GET /models (Returns Nabu internal format) - GET /v1/models (Returns standard OpenAI model list JSON footprint) - GET /tts/models - GET /v1/tts/models

Query by type: ?type=llm|tts|all