updated #research to better handle adding custom context to chat messages. lots of cleanup too,

🤖 updated #research project to better handle context windows and even an experimental autogpt-like commands

🤖 played around with adding extra context in some local Ollama models. Trying to test out some real-world tasks I'm tired of doing. #research

expanded #research project to handle messages and history better. Also added the ability to ask one-off prompts that may not get added to the history log.

🤖 updated #research app to better handle "continue" commands

Get email thread context into the AI writing prompt

🤖 added an inline keyboard and trying out some new options for my telegram bot #conf

Dynamically choose how many recent messages to include for context, based on total token lengths. Instead of just taking the recent 10 messages like I did before, which meant sometimes I wasn't using as much context as I had available, and other times I was exceeding the limit of tokens which broke ChatGPT #marcbot

Added UI for adding context items to the AI runner #saasco

🤖 worked on discord scheduling bot (more like inbox but bot works too) for #djangocon

#ollamachat Added support to persist conversations and multiple conversation with switching between models

updated story writer to use chatgpt and llama2 to smooth out some issues where large context windows would be helpful

Refactored the code so that everything uses a common AI message analyzer component instead of a copy with slight changes for Discord, Reddit, X, Telegram etc. Was starting to get unwieldly with 4 different integrations #watchdog !private

🤖 added more meta options to my code writing robot. #mylife

🤖 Wrote a git commit message bot using an LLM to create better commit messages #research

🤖 converted more management commands to telegram commands #conf

🤖 more robot work, but figured out how to get better results limiting num_predict for git commit messages #research

livecode replacing context with a shared xstate service #useauth

worked on persisting chat for logged in users #albertbro

🤖 Used Claude's Long context prompting tips guide to let it write a file contents to xml script to make it easier to compile a series of files in a project for Claude's projects and artifacts #research docs.anthropic.com/en/docs/bu…