GBrain v0.36.1 just landed on main. It ships something that was an idea from a hackathon team on Saturday: a team created Hindsight which track predictions over time to help you make better ones. This PR is inspired by that submission, and adds first-of-its-kind full evals.
Many users praised GBrain v0.36.1's hindsight calibration loop because it lets AI systems remember past mistakes and improve decision-making over time.
No Digg Deeper questions have been answered for this story yet.
#121Also 28 bug fixes landing in a single bug fix wave that rolls up 22 community PRs and 14 issues
GBrain v0.36.1 just landed on main. It ships something that was an idea from a hackathon team on Saturday: a team created Hindsight which track predictions over time to help you make better ones. This PR is inspired by that submission, and adds first-of-its-kind full evals.
GBrain is SOTA on LongMemEval out of all known OSS retrieval and memory repos without LLM query rewriting by more than 1%.
It's MIT license and I run it on my own 100k markdown page OpenClaw/Hermes Agent brain.
https://github.com/garrytan/gbrain
Over time your brain will not just tell you what people thought or said a year ago, but whether they were right.
You could see how this might be useful to me personally. :-)
A brain helps you and your agent think better. Epistemology meets track record.
Over time your brain will not just tell you what people thought or said a year ago, but whether they were right.
You could see how this might be useful to me personally. :-)
A brain helps you and your agent think better. Epistemology meets track record.
GBrain v0.36.1 just landed on main. It ships something that was an idea from a hackathon team on Saturday: a team created Hindsight which track predictions over time to help you make better ones. This PR is inspired by that submission, and adds first-of-its-kind full evals.
Full eval report and fixtures published and open source here https://github.com/garrytan/gbrain-evals/blob/main/docs/benchmarks/2026-05-18-brainbench-cat14-cat15-calibration.md
Any memory system is welcome to run theirs against these evals and I'll publish them alongside.
GBrain v0.36.1 just landed on main. It ships something that was an idea from a hackathon team on Saturday: a team created Hindsight which track predictions over time to help you make better ones. This PR is inspired by that submission, and adds first-of-its-kind full evals.
@jonrobflaig Daily GBrain Doctor task
@NostaIgicGareth @garrytan Do it
@garrytan Hi Garry, i like what you do with GBrain, great job, check this out, it’s my actual setup for personnal RAG, i’m sure you gonna feel it :
https://aichoes.lovable.app
@garrytan Hey bro, I’m going to tokenise this GBrain and send you all fees to your GitHub. This is awesome.
@NostaIgicGareth @garrytan Whoa 😳 @garrytan
@garrytan Can someone who isn’t a technical founder make use of GBrain?
@garrytan Nice! Calibration loop + personal track record is huge for better thinking. Great hackathon inspo and open evals will check it out.
@garrytan Here you go https://github.com/jackccrawford/geniuz
@TheTanjiro69426 @garrytan C8TSwcUMavmSQY8dozXWyy9DhD5QBCT41PBdRRAZpump
I just implemented the previous version into a drayage TMS that mixes structured data with conversational knowledge. It monitors data streams from Slack, email, text messages, and phone calls to understand how freight moves within my company. Thank you, @garrytan, this is a great solution for optimizing trucking company operations.
@garrytan how do you handle persistent failure modes?
@garrytan tracking your own predictions sounds humbling as hell tbh
@garrytan How do you track performance when the Doctor task finds/addresses a failure mode? Our framework has an actual performance log & 3 strike termination policy where incumbents learn from previous terminations. Curious if I'm over-engineering anything..
@garrytan it is fantastic
recoverable context is king
@garrytan yeah this is the part i want more of. not just a memory pile, but a little scar tissue from what actually happened after the prediction
@garrytan Does it have decay logic? One of the biggest problems with AI memory that it remembers things that I’ve mentioned a year ago that is no longer relevant. Will test and post results anyways
