Block a user
2f9b00ae29
Have settled on testing and using the Qwen2.5:1.5b model as it plays much nicer with the JSON requirements - way better than the Gemma models did. Have refined the calls via LlmManager.kt and MediaPipeAdapter.kt and tested using battery, ping and a new weather skill. Half way through updating further to allow for SMS reading and updating the SkillManager.kt again to accommodate this with a two-pass bridge for python calling.
8ba9cb7a62
Check in part way through the engine update to MediaPipe. Have the system working and now troubleshooting the LLM models and sizes and how they handle the JSON and tool calling loops...Check in to allow dev on the laptop as well from here.
d4322740e2
Currently confirmed offline model download and communications are working. Next step is to build the JSON schema for local models to understand tool calling
93a2c48e4b
Made the decision to initially stick to the Ollama LLM backend and not go down the whole c++ download and compile for MLC LLM usage. Update to the UI to now include the model downloaded for the Llamatik backend to use. Tested as working and downloading models successfully.
66e4b04ed8
Updated the AliceAgent.kt filename and refactored - hopefully haven't broken anything else.....
25167ff6cb
Large changes to the UI ( and matching backend) to allow for chat threads, system prompt changes, and starting the MLC LLM integration options for the backend AI switch/choice
8b474d81be
UI and Skills settings update to now allow for on the fly model changes and the skills directory to be changed. Both confirmed as working
1a3d6bde7c
Added the langchain4j and LLM routing functionality. Tested with a basic ping_server skill which worked.
12c54e237b
Updated to now auto create the skills folder ready for the skills plugin. Also updated the UI to show this path and allow for "real" navigation between settings and the chat interface.
0340f80adb
Implemented a test remote callout to an LLM running on the local host machine for development and testing purposes. Successfully getting LLM responses in the Alice messages UI
7c1bc79fb2
Now have the message box - send to langchain4j routine set up and confirmed, even without an LLM to test against so we prove that the onscreen workflow is OK
f96189a509
Updated the processAudioLoop to mke sure that the buffer was always flushed after an output stream so we didn't miss any characters. Also made many modifications to the python script running on the Pico to ensure that we get dual inputs and neater outputs now.
75b63f91ea
Updated to continually check for the a valid USB connection on each press of the mircophone and also totally rewrote the send to Pico code as well to me more robust...
9f6d67a567
Added the debug.keystore to the repo so that I can move between devices and build the app without getting installation errors on the phone.