Add developer api docs using redocly because there are multiple customers using this now and I want to stop forwarding telegram DMs with docs Too bad the UI looks kind of meh, may look for an alternative or try to theme it later #watchdog

Ship start and end date filtering for the community statistics developer API h/t @marc #watchdog

Make analytics page design more uniform, now to hook up data to the metrics table #watchdog

Start on analytics page redesign to include shareable, global statistics for all communities that a customer has #watchdog

Close new customer at $59/mo over X DM because he noticed the #buildinpublic logo on the landing page and runs a similar community on X #watchdog

Get nice feedback from influencer #watchdog

Pay YouTube influencer with 10K subscribers $100 as a marketing experiment for a #watchdog tutorial video + promotion on their Discord server

Move static guidelines to the the top of the prompt to try to take advantage of prompt caching for lower costs #watchdog

Ship quick fix to block customers from trying to supply youtube videos as Q&A sources #watchdog

Mark image read errors as retryable so that the retry mechanism I implemented actually works 🤦‍♂️ #watchdog

Negotiate with influencer with 12K followers for a YT video for #watchdog who seems excited about the product and agree on $100 as a test budget

Auto retry image download requests for X communities because sometimes their server times out #watchdog

See acceptable latency increase and 50% cost reduction (!) after enabling service_tier = flex #watchdog

Try service_tier = flex to cut costs w/o sacrificing quality because I'm getting killed on OpenAI spend #watchdog

See a ~3x cost reduction moving from Gemini 2.5 Pro to OpenAI O3 with near identical quality metrics Costs are still too high and will need to optimize further to maintain a reasonable profit margin. My AI credits have finally run out and I'm paying bills like a real business owner 😅 Now trying an experiment with the "flex" processing tier to see what latency looks like. Need to cut costs further while maintaining quality #watchdog

Swap AI models and start monitoring to see how latency and spend improves if at all #watchdog

Inspect current P99/P50 message analysis latency after 24h and it's pretty shit, let's see if the new model makes a dent when I enable it in a few minutes #watchdog

Ship message analysis latency metrics so I can see if performance improves when I switch AI models #watchdog

Add "Start a Trial" link to the sign-in page for #watchdog to match #downloadpage's login screen

Manually outbound to unhappy customer of competitor's bot and ask them to use #watchdog