HackerTrans
TopNewTrendsCommentsPastAskShowJobs

randall

no profile record

Submissions

Show HN: Gambit, an open-source agent harness for building reliable AI agents

github.com
91 points·by randall·6 mesi fa·27 comments

comments

randall
·27 giorni fa·discuss
[dead]
randall
·3 mesi fa·discuss
this isn't strictly true. not that it thinks, but it can reason about the tokens that led to the outcome.
randall
·3 mesi fa·discuss
this is "you ship your org chart" not ai.

https://en.wikipedia.org/wiki/Conway%27s_law
randall
·3 mesi fa·discuss
Does anyone have a link to the "read more"?
randall
·3 mesi fa·discuss
exactly. They probably have unsustainable margins on accident.
randall
·4 mesi fa·discuss
sweet.

I've been out of the game for a bit but it's great to hear.
randall
·4 mesi fa·discuss
This is the least informed take i've ever seen.

I worked at fb, and I'm 100% certain we sponsored VLC and OBS at the time. It would be strange if we didn't sponsor FFMPEG, but regardless (as the article says) we definitely got out of our internal fork and upstreamed a lot of the changes.

I worked on live, and everyone in the entire org worships ffmpeg.
randall
·4 mesi fa·discuss
even marvel villain thanos was given a plausible motivation.

i don’t think if you actually knew him as a person you’d think this was a reasonable perception.
randall
·4 mesi fa·discuss
same. this is about losing a negotiation and saving face / exacting revenge.
randall
·4 mesi fa·discuss
i don’t get it. what’s his motive in your view? he literally has no shares in openai.
randall
·5 mesi fa·discuss
I wanted the AOL one to say "Welcome" before "you've got mail!" lol
randall
·5 mesi fa·discuss
The last 3 years of LLM progress, to me, feel like 1994-1998.
randall
·5 mesi fa·discuss
this is like irl cryptographic signatures for content lol
randall
·6 mesi fa·discuss
thanks for the PR! :)
randall
·6 mesi fa·discuss
yaaaaa exactly. You're totally on the same wavelength as me. Let's be friends lol
randall
·6 mesi fa·discuss
once your team comes to a consensus on what PII is, you can roughly guarantee it... especially as models improve.
randall
·6 mesi fa·discuss
So two things.

1/ crypto signing is totally the right way to think about this. 2/ I'm limiting prompt injection by using chain of command: https://model-spec.openai.com/2025-12-18.html#chain_of_comma...

we have a "gambit_init" tool call that is synthetically injected into every call which has the context. Because it's the result of a tool call, it gets injected into layer 6 of the chain of command, so it's less likely to be subject to prompt injections.

Also, relatedly, yes i have thought EXTREMELY deeply about cryptographic primitives to replace HTTP with peer-to-peer webs of trust as the primary units of compute and information.

Imagine being able to authenticate the source of an image using "private blockchains" ala holepunch's hypercore.
randall
·6 mesi fa·discuss
i have a huge theory here that idk when we’ll implement but it has to do with “quorums” and other stuff.

hard to explain… we’ll keep going.
randall
·6 mesi fa·discuss
right now yeah we’re just dropping context… sub agents are short lived.

thinking about ways to deal with that but we haven’t yet done it.
randall
·6 mesi fa·discuss
yeah the way to do that stuff is through zod schemas… input and output schemas.

you can set up really complex validation.

thanks for checking it out!!