I use a bunch of different coding agents but they always have annoying permission pop-ups or not the right kind of sandboxing. I wanted something simple and flexible so I made it over the weekend.
You use it like so:
- Get the binary in whichever way you like (you can use Nix, Mise or just get it through the releases page).
- Configure via `.agent-run/config.toml` to make the folders you care about read-write, and add any environment variables.
- Run your agent like so: `agent-run <pi|claude|codex|opencode>`.
It's very generic so technically you can use this for any command.
Currently only supports Linux because I don't have access to a macOS machine and sandbox-exec looks daunting.
Every other piece of
* the old tier-1 dispatch JIT (the ~1500 lines of per-opcode helpers,
* the jump-table dispatcher, the jit_rt_* runtime shims) was deleted
* because benchmarks showed tier-2 dominated on every workload that
* reached the JIT at all. */
No human cares about including such an irrelevant detail as the lines of code in helper functions. Obvious LLM context spew, obvious AI slop project. Please stop posting things you didn't bother actually working on.
I mean, if government overreach (IP, DMCA 1201) is preventing us from using the things we pay money for in any way we might, might as well add more government overreach on top to claw some rights back?
I've personally driven adoption of Bazel across two teams in our company over the last two years (we have an extremely fragmented set of repos that touches everything from hardware, to end-to-end simulation, to cloud software, to CDK stacks that deploy the software). There are still some aspects (ha) that confuse me like transitions, exec groups and constraints, but most of the time it's all about building a graph of actions (command executions) by executing code, and hooking up external dependencies into the graph inputs. Mind telling what confuses you in particular? Maybe I can be of help somehow.
It is under a license that prevents using the source in any way you choose:
You may not use the klaw source code to operate a multi-tenant managed
service (Software as a Service) where the primary value proposition is
providing AI agent orchestration capabilities to third parties, unless
you have obtained explicit written authorization from each::labs.
It's source-available, which is cool, but don't muddy the waters.
Call it larping, being performative etc. but it is a concept as old as time. People emulate the interface of successful people without actually having the implementation of successful people.
> Ignoring requires-python upper bounds. When a package says it requires python<4.0, uv ignores the upper bound and only checks the lower. This reduces resolver backtracking dramatically since upper bounds are almost always wrong. Packages declare python<4.0 because they haven’t tested on Python 4, not because they’ll actually break. The constraint is defensive, not predictive.
This is clearly LLM-generated and the other bullet points have the same smell. Please use your own words.
Cross-compiling doesn't work because you're not defining your dependencies correctly and relying on the existence of things like system libraries and libc. Use `zig cc` with Go which will let you compile against a stub Glibc, or go all the way and use a hermetic build system (you should do this always anyhow).
For the GC it sounds like they don’t have generations which means copying long-lived objects needlessly due to the generational hypothesis. Interesting idea with the mailbox allocator, but how do these two allocators interact? Is the heap non-regional, or are they allocating into separate regions?
I would very much like for him not to ignore the negativity, given that, you know, they are breaking the entire fucking Internet every time something like this happens.
A lot of claims about being "privacy first", but is there any way to actually verify these claims? For example they claim "no logs", but unless I log into their servers and personally check there is no way I can be sure, right? Is there something I'm missing?
You use it like so:
- Get the binary in whichever way you like (you can use Nix, Mise or just get it through the releases page).
- Configure via `.agent-run/config.toml` to make the folders you care about read-write, and add any environment variables.
- Run your agent like so: `agent-run <pi|claude|codex|opencode>`.
It's very generic so technically you can use this for any command.
Currently only supports Linux because I don't have access to a macOS machine and sandbox-exec looks daunting.