feat: refined Truncation Strategy to save significant amount of tokens #203
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I analyzed the truncation strategy that we are using for summarizing the message history for the LLM.
Currently, we are keeping all of the screenshots in the history and send them to the LLM with every request.
As you can imagine, this costs tons of token with a minimal chance of providing any relevant information to the LLM.
Hence, I implemented a new truncation strategy, that will only send the latest screenshot in the history to the LLM and replace all other screenshots with a string („Screenshot removed to save tokens“).
In small experiments, this allowed me to save up to 90% of token consumption.
However, there might be scenarios where keeping previous images in the history might help to explain e.g. when searching for something or trying to understand the UI transition graph. Hence, I marked this truncation strategy as experimental.
Summary of what I added: