kamal

Serial employee. 2nd engineering hire at @envoy. Prev: @viki (acq). 🇲🇾

[kamal @kamal]

The Kamal stack is

K - you don't have
A - product market fit yet
M - why you fucking around
A - with kubernetes
L - just use heroku

[kamal @kamal]

@thorstenball @evanphx Do you find yourself pushing to a short lived branch so that you can archive the thread and have fresh new context with a new thread? For me in spaces I have very few threads and almost no archived because it’s like working memory

[kamal @kamal]

@hameonline @sudo_goreng @AmpCode wait a min. this whole time i was basing the price calculations on https://t.co/FgMy5LItZi

[kamal @kamal]

@sqs @maxjendrall @AmpCode the thread storage sounds really intriguing. also my security person is terrified of mcp credentials and secrets

[kamal @kamal]

Asked Opus to speed up 30+ amd64-only containers on Asahi Linux (M1 Max MBP). Instead of tuning qemu, it took the images apart: kept jar/Ruby/Node layers, replayed Dockerfiles on arm64 bases, rebuilt native gems/addons. API calls: ~1s → ~0.2s. Didn't know you could do that!

[kamal @kamal]

@sqs 11 and then 10. For 10 specifically, I want the ability to restrict Raw Models selectable by workspace users.

[kamal @kamal]

@sqs @AmpCode Do you publish egress IPs?

[kamal @kamal]

@AmpCode @sqs if my LLM gateway is in my tailnet, can Amp reach it? My understanding is only the orb is a Tailscale client but the agent loop lives outside of it

[kamal @kamal]

@AmpCode do you have a published list of egress IPs? i'd like to lock down my llm gateway given *gestures* .. all this

[kamal @kamal]

Testing out self-hosted orbs on @flydotio . currently custom orbs only have a single size. the workaround is to run multiple providers. @AmpCode will you pass in the size so that we can just have a single provider that can vend different-sized orbs?

[kamal @kamal]

Did @AmpCode removed the oracle in the mode dial?

[kamal @kamal]

@ajhodls fundraising

[kamal @kamal]

software factories and cloud-based harnesses have a lot of overlap. what makes a factory like @WorkOS atlas different from @AmpCode?

[kamal @kamal]

@thorstenball @evanphx @AmpCode Let’s be real. The only reason to get the duo https://t.co/XgqfA9CI6A

[kamal @kamal]

@thorstenball Morb on iOS?

[kamal @kamal]

@dwr That’s Google cc https://t.co/u4K2IcEP91

[kamal @kamal]

this was my biggest pet peeve with graphite

[Sameen Karim @sameenkarim]

gh stack CLI now supports git worktrees 🥞
You can manage stacks where each branch is in a separate worktree, while still taking advantages of things like cascading stack rebases

https://t.co/qpsu1hunB3

[kamal @kamal]

This is confusing. Just got this error message:

The 'gpt-6.1-sol' model is not supported when using Codex with a ChatGPT account

[kamal @kamal]

@thsottiaux (Ultra)Fastpass. Like the Mario star. 60 mins of ultrafast at normal rates.

[kamal @kamal]

Do all the people posting impressive motion design videos made by opus 5.5 realize that it was post-trained on “eval packs” that had these styles labeled as good? It’s a huge business to sell these to labs. Imagine the mercor for high taste evals. LoveFrom would make a killing

[kamal @kamal]

@kylegalbraith @depotdev Small correction: it was indeed GHA runners because it needed a macOS machine for an iOS build. Depot CI doesn’t seem to have the option to run on a Mac?

[kamal @kamal]

i'm benchmarking @depotdev CI against github actions. marginally faster when running but the queue times are quite a bit longer than github's.