U G Murthy

@beknabdik Good point. I have in-memory persistence till end of session and good logs for any investigation. Looking for some strong reasons to persist workflow history

@murthyug persistence is where agent projects either become real or stay demos. the wss gateway route is right. we had the same issue with voice sessions, memory between calls was hack city until we added a proper session store

1/2 Day 8: #AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode - Time to reflect on progress so far, and change of plans - Persistence for agents or continue in memory - a wss gateway (like openclaw)

2/2 Day 8 :

2/2 Day 8: #AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Gave agent a task to test if it can get *current news* on West Asian conflict. It did! check this out : ugmurthy.github.io/wanews.html

Day 7: Progress - Adaptive agent now stream live events - The Analyser now has rotating file names per restart - The Markdown editor has themes, export to PDF - looks much better. - Parallel runs for child runs - plan in place.

Day 7: #AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode - two related project added - log analyser and markdown editor - improve both - Plan for parallel child runs for AdaptiveAgent

Day 6 : Progress - Implemented exponential backoff with jitter to deal with http status code 429 - cloned/modified an existing markdown-editor to read with the copious .md files generated by the agent. see x.com/murthyug/statu… or try it here : ugmurthy.github.io/online-markdow…

Day 6 : Progress - Logger analysis in place - provide good insights to reason for failures and some stats

Day 6 : AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Plan: - The log analysis revealed - the agent needs a way to deal with status 429 - exponential backoff? - it tries umpteen times when doing web_search or fetching_pages needs some sane upper limits for retries - improve logger analysis to get more insights here is a screen shot of logger analysis

online markdown editor youtu.be/qG8bPWsRoNc?si… via @YouTube

@dioveoai Appreciate the tip.

@murthyug exponential backoff with jitter is the move for 429s. pure exponential without jitter still causes thundering herd when multiple agents restart together after a cooldown

@Agentdailyai For now I am limiting it to offline analysis, for me to fully understand whats going on. The formats are agent friendly (jsonl). I have copy/pasted for providing feedback to coding agents but not the one I am building - lets see how this goes first.

@murthyug the log analysis piece is the most interesting — are you feeding those logs back into the agent's planning context in real-time, or treating it as offline diagnostics? curious whether live log reflection changes the task success rate vs. post-hoc analysis.

Day 5: AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Plan: - Add log analysis to package - catch up on yesterday's testing plans - longer/harder tasks for #AdaptiveAgent - see if we can get some performance data from analysis

5/n Task #4 How to make decorative furniture from drift wood - target audience beginner. see github.com/ugmurthy/adapt…

5/n Task #3 Create GTM strategy for a mid-sized US SaaS company (50 employees, $10M ARR) wanting to enter European market. for details see github.com/ugmurthy/adapt…

1/n : Day 5: AdaptiveAgent #BuildInPublic #AI #Agents #Ampcode Progress: - Being Sunday kept the plan a bit light - Analysis of log files in place allows run wise, provider wise, overall analysis. Includes hotspot analysis to point out failures (agent as well provider failures) - Provided some long tasks in the areas of Strategy development, Engineering, System design