Uploaded April 2026 | Updated November 2026, 2 days ago
Summary of Content
The video follows your interaction with the on-device AI models, specifically Gemma 4, focusing on the limitations and bugs you encountered:
1. Feature Failures: You noted that the Agent Skills section is currently unusable, as it crashes frequently.
2. Lack of Multimodality: You highlighted a significant flaw in the AI Chat mode: the inability to import external files like Google Docs, audio, or images.
3. History & Memory Issues:
* The "Input history" appears to leak from other modes (e.g., showing mood logging attempts that you mentioned had failed).
* When reloading a previous prompt (specifically "tell me how to practice CBT on myself"), the model regenerates the response from scratch instead of resuming or remembering the previous
interaction.
4. Performance Inefficiency: You expressed frustration that the app uses an "enormous amount of CPU/GPU" to regenerate static content rather than maintaining session memory.
Analysis of Key Data
The video reveals several specific entries in your Text input history that provide context for your day (April 14, 2026):
* Mood Corrections: One entry notes a "terrible day score of 1" due to feeling unrested, but a later correction updates the score to 7 after an early swim.
* CBT Interest: You are using the on-device AI to explore Cognitive Behavioral Therapy (CBT) techniques, specifically "Phase 1: Becoming a Thought Detective."
* Relationship Context: The history mentions "Gottman" (likely referring to the Gottman Method for couples therapy) and practicing being a better listener.
The video serves as a technical bug report on the Edge Gallery's lack of state management and multimodal support.
Summary of Content
The video follows your interaction with the on-device AI models, specifically Gemma 4, focusing on the limitations and bugs you encountered:
1. Feature Failures: You noted that the Agent Skills section is currently unusable, as it crashes frequently.
2. Lack of Multimodality: You highlighted a significant flaw in the AI Chat mode: the inability to import external files like Google Docs, audio, or images.
3. History & Memory Issues:
* The "Input history" appears to leak from other modes (e.g., showing mood logging attempts that you mentioned had failed).
* When reloading a previous prompt (specifically "tell me how to practice CBT on myself"), the model regenerates the response from scratch instead of resuming or remembering the previous
interaction.
4. Performance Inefficiency: You expressed frustration that the app uses an "enormous amount of CPU/GPU" to regenerate static content rather than maintaining session memory.
Analysis of Key Data
The video reveals several specific entries in your Text input history that provide context for your day (April 14, 2026):
* Mood Corrections: One entry notes a "terrible day score of 1" due to feeling unrested, but a later correction updates the score to 7 after an early swim.
* CBT Interest: You are using the on-device AI to explore Cognitive Behavioral Therapy (CBT) techniques, specifically "Phase 1: Becoming a Thought Detective."
* Relationship Context: The history mentions "Gottman" (likely referring to the Gottman Method for couples therapy) and practicing being a better listener.
The video serves as a technical bug report on the Edge Gallery's lack of state management and multimodal support.





![gaia AMD fails Recording made 2026 05 10 103334
Bug Report: Gaia Chat Model Mismatch & Lemonade Network Timeout
Target Repository: lemonade-server / gaia-chat
Environment:
OS: Windows (PowerShell 7.6.1)
Server: Lemonade-server (v. 2026/latest)
CLI: Gaia CLI
1. The Issue
There is a discrepancy between the model loaded via the lemonade CLI and the model actually invoked by gaia chat. Furthermore, when gaia chat attempts to process a query, it triggers a LemonadeNetworkError (CURL timeout), despite the model being successfully loaded in the server.
2. Steps to Reproduce
Load specific model via Lemonade:
lemonade load Qwen3-VL-4B-Instruct-GGUF
(Result: Model loaded successfully!)
Run Gaia Chat with local index:
gaia chat index [path_to_pdf] query what is this document about? debug
Observe Model Output: Gaia defaults to Gemma-4-E4B-it-GGUF instead of the loaded Qwen3.
Observe Network Failure: The process hangs and returns a CURL error: Timeout was reached.
3. Observed Behavior
Model Inconsistency: Even though Qwen3 was loaded, Gaia attempted to use Gemma-4.
Network Timeout: gaia.llm.providers.lemonade.chat logs a LemonadeNetworkError.
Deprecation Warning: The legacy lemonade-server command is still active but throws argument errors, necessitating the use of the new lemonade load syntax.
4. Expected Behavior
gaia chat should detect and use the currently active model in the lemonade-server or provide a flag to specify the local model to prevent falling back to a default that may not be active/cached, which likely leads to the timeout.
5. Relevant Logs
Plaintext
[DEBUG] Model: Gemma-4-E4B-it-GGUF (Expected: Qwen3-VL-4B-Instruct-GGUF)
[WARNING] Lemonade error: type=LemonadeNetworkError payload={error: {message: Network error: CURL error: Timeout was reached}}
[ERROR] Error in send_messages: Couldnt reach the local LLM. gaia AMD fails Recording made 2026 05 10 103334](https://i.ytimg.com/vi/ts_mRvJzIMo/mqdefault.jpg)




