TOTAL RECALL Generates a summary for an old text saved at left panel (chats) into the annotation at your chat.
To run the plugin-generator, select it from the list of models (turn it on at the drop-down top-list ) ;
then type below in the field of chat the command summa: and part of chat name from the list on the left panel.
In the config, the list of commands could be expanded using a separator(;). Add your own command. Example:/summa;
All words are synonyms for a single command that is a prefix for the name of the chat. Any languages are supported. It's up to you.
Other words are not processed at all, they just print short instruction right in the chat.
Plugin available at https://lmstudio.ai/tupik in updated version.
lms get tupik/summary-chat
After installation the plugin doesn't appear in the right panel. It disappears, which is expected. This is how all generators work. (a type of plugin) Only after launching from the central drop-down menu it does the configuration panel appear on the right one.
Total Recall -- How it works:Chat Folder Name from drop-down list if you have folder.enter command and part of an old chat name from left panel.summa: command. You can enter your custom command in the config.LLM from the field Model-name no active in the config. Empty field means the current model is used.
With empty select the plugin-generator uses any LLM model that is already loaded.
Added the option to select a folder, since files with the same names can exist in different folders.
List if folders stored/caches in the file conversations-cache.json
The source files have now been consolidated. There are only 2 files /src/index.ts and src/generator.ts.
Sometimes output shows Tokens:0. This isn't an error; this value is stored within the chat file.
The program itself calculates tokens when it needs to. And we suddenly intervene in this flow.
The chat files flow aren't still works in the way we might expect. Especially old, preserved ones.
Summarizing System prompt now in file summa-prompt.md (Maybe we should make several commands for different prompts?)
Best models for this - with Instruct/it capability (they follow the prompt better).
The developers at Alibaba (the creators of Qwen) specifically trained the model to recognize the strings /think and /no_think.
When the tokenizer sees this trigger, it instantly replaces the model's internal state, generating the closing tag,
telling the AI-network: "reasoning is over, go straight to the final answer." Where else does this work? Only in the Qwen3 family (and its derivatives).
TOTAL RECALL Generates a summary for an old text saved at left panel (chats) into the annotation at your chat.
To run the plugin-generator, select it from the list of models (turn it on at the drop-down top-list ) ;
then type below in the field of chat the command summa: and part of chat name from the list on the left panel.
In the config, the list of commands could be expanded using a separator(;). Add your own command. Example:/summa;
All words are synonyms for a single command that is a prefix for the name of the chat. Any languages are supported. It's up to you.
Other words are not processed at all, they just print short instruction right in the chat.
Plugin available at https://lmstudio.ai/tupik in updated version.
lms get tupik/summary-chat
After installation the plugin doesn't appear in the right panel. It disappears, which is expected. This is how all generators work. (a type of plugin) Only after launching from the central drop-down menu it does the configuration panel appear on the right one.
Total Recall -- How it works:Chat Folder Name from drop-down list if you have folder.enter command and part of an old chat name from left panel.summa: command. You can enter your custom command in the config.LLM from the field Model-name no active in the config. Empty field means the current model is used.
With empty select the plugin-generator uses any LLM model that is already loaded.
Added the option to select a folder, since files with the same names can exist in different folders.
List if folders stored/caches in the file conversations-cache.json
The source files have now been consolidated. There are only 2 files /src/index.ts and src/generator.ts.
Sometimes output shows Tokens:0. This isn't an error; this value is stored within the chat file.
The program itself calculates tokens when it needs to. And we suddenly intervene in this flow.
The chat files flow aren't still works in the way we might expect. Especially old, preserved ones.
Summarizing System prompt now in file summa-prompt.md (Maybe we should make several commands for different prompts?)
Best models for this - with Instruct/it capability (they follow the prompt better).
The developers at Alibaba (the creators of Qwen) specifically trained the model to recognize the strings /think and /no_think.
When the tokenizer sees this trigger, it instantly replaces the model's internal state, generating the closing tag,
telling the AI-network: "reasoning is over, go straight to the final answer." Where else does this work? Only in the Qwen3 family (and its derivatives).