ai-christianson

What do you think the problem is?

Gobii did always-on, connected to comms channels (sms, email), full computer/headed browser 8 months ago, in a much more secure and k8s-native way.

Control of where data goes is always an option. People just need to make that choice.

Not sure what gives you that idea. One of our superpowers is that we're MIT licensed and deployable to private clouds, or even fully airgapped with 196gb+ of vram to run minimax on vllm + Gobii.

For anyone looking at alternatives in this space - I built Gobii (https://gobii.ai) 8 months before OpenClaw existed. MIT licensed, cloud native, gVisor sandboxed.

The sandboxing part matters more than people think. Giving an LLM a browser with full network access and no isolation is a real security problem that most projects in this space hand-wave away.

Multi-provider LLM support (OpenAI, Anthropic, DeepSeek, open-weight models via vLLM). In production with paying customers.

Happy to answer architecture questions.

If you run it for an hour, yes.

If you want to go really minimal you can do raw git+ssh and hooks (pre/post commit, etc).

We went down the WASM sandboxing rabbit hole at Gobii when building our agent infra. The pitch is appealing until you realize the tradeoff: you either accept a limited environment with reimplemented tools, or you emulate your way back to a full Linux system (like agentvm does at 173MB) and wonder why you didn't just start with gvisor or Firecracker.

We landed on gvisor in k8s. Our agents run headless Chromium for browser automation, ffmpeg for media processing, yt-dlp, ripgrep, fzf - real tools that would be a nightmare to port or reimplement. Actual Linux with the full ecosystem, solid isolation, no emulation overhead.

Interesting project though - the capability-based tool validation layer seems useful regardless of what's running underneath.

They say compressed... but isn't this just "minified"?

They might actually be going all-in on a robotaxi future, i.e. betting on a future where car-ownership is not the default.

I’m Andrew I. Christianson (co-author) from Gobii. This post is the production sandbox we built for running untrusted agent workloads: per-agent isolation (gVisor), default-deny egress with proxy-only outbound, deterministic filespace sync, and audit logs for every tool call.

Happy to answer anything, especially threat model edge cases and failure modes. Code links are in the post if you want to go straight to implementation.

Probably even more so if it can fully fit into CPU cache.

Yeah, the harness quality matters a lot. We're seeing the same pattern at Gobii - started building browser-native agents and quickly realized most of the interesting workflows aren't "code this feature" but "navigate this nightmare enterprise SaaS and do the thing I actually need done." The gap between what devs use Claude Code for vs. what everyone else needs is mostly just the interface.

Then there's the shuffling around of atoms.

Gobii AI | Senior Software Engineer | REMOTE (US) | gobii.ai

Gobii is the easiest way to deploy browser-use agents in the cloud or on-premises. We're building an open-core AI workforce platform—browser-native agents and automation tools. We have paying customers.

The entire team contributes to browser-use. We also have contributors to Apache NiFi, LiteLLM, and RA.Aid (open-source coding agent I created). Open source is core to how we build and how we think. We're backed by Open Core Ventures.

HQ is San Francisco, but much of the team is in the DC area. East Coast timezone preferred, DC area even better—but we're fully remote.

What we're building: The orchestration layer for browser-use and AI agents. Systems that navigate, reason, and execute on the real web.

Who we're looking for: You've built agentic systems that shipped to production. You have opinions about tool use, context management, and why most "agents" are just glorified prompt chains. You strongly prefer working in the open—a significant part of this role is contributing to OSS repos. We strongly prefer candidates with existing open source contributions.

Why now: We have strong traction and are expanding the team. You'd be early—with real influence over product, architecture, and how we work. High autonomy, high trust, small team of people who know what they're doing.

Comp: Cash + healthcare + modest 401k match + options. In line with SF-based VC-backed startups at pre-seed stage.

Stack: TypeScript, Python, browser automation, LLM orchestration.

Email: careers@gobii.ai (subject: "HN")

"GNU Taler is a Free Software payment system that preserves the privacy of payers while ensuring that income is visible to authorities."

I get that we're aiming to be compliant here, but this seems to run very contrary to GNU ideals of freedom and privacy.

Having a hobby kind of like going shallow into one or more topics.

Wouldn't the best way to spend a life be to go all in on one thing?

It's not like a video game where we can just try again. If the main quest isn't right, better find a new one, not just add distracting side quests.

I've had really good luck with https://github.com/SolidCode/SolidPython

Is OpenSCAD still being maintained?

I love garage. I think it has applications beyond the standard self host s3 alternative.

It's a really cool system for hyper converged architecture where storage requests can pull data from the local machine and only hit the network when needed.

Or maybe we just move away from cloudflare-like services altogether.

Sorry bro we're going for functionality over beauty atm.

Great idea :)

Agreed, the emergent behavior is the most interesting and valuable part. We don't want bad emergent behavior (agents going rogue) but we do want the good kind (solving problems in unexpected ways.)

This is quite literally what we've built @ Gobii, but it's prod ready and scalable.

The idea is you spin up a team of agents, they're always on, they can talk to one another, and you and your team can interact with them via email, sms, slack, discord, etc.

Disclaimer: founder

That makes it seem like we're all in someone else's "dwarf fortress" game :)

I'm mainly a Linux user, but recently switched to a M4 mbp for video editing (davinci resolve) and local LLMs.

Is Asahi able to run mlx with the full Apple hw optimizations?

I'm guessing it's a long shot for Resolve to run there, let alone with hardware optimizations.

There's massive hardware and energy infra built out going on. None of that is specialized to run only transformers at this point, so wouldn't that create a huge incentive to find newer and better architectures to get the most out of all this hardware and energy infra?

Somewhere, a function just called itself and raised a seed round.

Isn't there a link between high intelligence and depression as well?