Ornith was fine for basics but lost coherence on day-long projects. But GLM 5.3 Flash on twin Sparks is everything I hoped it would be. Thanks @MiaAI_lab and team for the recipes that accelerate the local near-frontier.

Mine saves every message I send in a file. Just before compaction it triggers /new & re-sends the relevant msgs, newest first, with a log of todos, work already done & the project charter. Same for subagents. This preempts context compaction, making it a non-issue. Game changer.

12 satellites loaded successfully

They had lots of test campaigns on Reddit in 2013-2015

Weekdays: gpt6 with Hermes. Weekends: Claude to audit, overhaul and improve the Hermes setup. Repeat.

This'll be fun

We gave human rights and personhood to corporations, so why draw the line at thinking sand?

Do you use TurboQuant in your recipes?

Autoresearch loop to map the fly connectome then (eventually) build an API to use it as a tiny AI engine: 1. Test for 1hr 2. Evaluate results 3. Consider direction changes based on the eval 4. Propose a new plan 5. Update wiki with test results, then 6. Loop and start over

2x 3Us on UARX OSSIE today, 4x 3Us on D-Orbit ION. Big day for ESI :)

@TheSpaceportCo is working on this

s/"dev team"/"Jev team"

Ha. Thanks Chris! One more will be ready to ship in four weeks if you want it. Might be the last one for a year while we focus on other big developments.

11 satellites loaded successfully

Today we’re loading the 11th sat of the year on customer unit #2 of our xe/kr loading cart. Happy to chat if you need a cart.

🫡 Up before dawn to fill another big satellite with xenon

Thanks Kyle! I'll take a look