Runta

Exclusive: Andreessen Horowitz backed Runta, a startup building sandboxes and guardrails to keep AI agents under control. Read more: thein.fo/4yuP02c

@martin_casado @JeffDean @drfeifei @alighodsi @Thom_Wolf @steph_palazzolo @runta @theinformation Thrilled to be partner with @martin_casado. Few people have thought harder about what happens when the infrastructure layer shifts. a16z.com/announcement/i…

@guanlan @runta @a16z Congrats Guanlan! Very excited for you!

Runta founder, Guanlan Dai on why every current AI safety approach is happening at the wrong layer: "Our existing infrastructure is mostly building for governing software. Software is something you can govern and constrain when writing. But agents are different. Agents are probabilistic. Whenever an agent makes a decision, it may make it on the fly, so the traditional way to govern software is no longer valid." "GPU and LLM convert watts into tokens, and Runta helps you convert the token to production value." "On the gateway level, or any of the in-between level, you only see the final result, final package. But you have no idea what exactly the agent is doing, why they are making this call, why they are building this package." "Inside the agent they have gone through a lot of steps. They have done tool calls. They might use budgets. They might access production data. That's the piece that's actually missing." "You cannot govern an agent here. You have to govern it underneath to see exactly what's going on at OS level." @guanlan @runta

@istdrc @guanlan @a16z Raft + Runta will be a powerful combo!

@guanlan @runta @a16z congrats!!

Today we're announcing @runta's $20M seed, led by @a16z. Software is constrained when you write it. Agents have to be constrained while they run. Runta is the execution layer that controls what AI agents can actually do.

We've been pushing existing compute tech into service for agents (VMs, containers, etc.). But none of it is really the right fit. Super excited to be working with @runta to take a fresh approach to the problem from one of the top systems teams on the planet. a16z.com/announcement/i…

We're thrilled to lead Runta's seed round. AI models have required rethinking the GPU stack. And now agents are doing the same thing for CPUs. @runta is rebuilding the execution layer agents actually need - a computer that's extremely efficient, runs locally or in the cloud, with the security and policy controls to run agents safely in the enterprise. Or anywhere. We've known @guanlan for nearly a decade, since his days leading Cloudflare's edge proxy and Kong's core proxy. @martin_casado is joining the board. Congrats to Guanlan and the Runta team. By @martin_casado @stuffyokodraws @appenz

Huge congrats @guanlan on the public launch! Our team has been using Runta for some of our RL environment task rollouts thanks to the early access, great experience. Super bullish on what Guanlan and the team are building! On a personal note, I’m also so proud to be a humble angel. This is actually my first angel investment in my life!

Apply for access here: dashboard.runta.com/login

If you're trying to give agents real work and control is what's stopping you, apply for early access today. Every new account starts with $50 in free credits, enough to run an agent for a month, with no credit card needed.

@NousResearch clawshell.org Get started: npm install -g @clawshell/clawshell sudo clawshell onboard

@NousResearch Release note: github.com/clawshell/claw…

ClawShell v0.2.0 is out. The headline is support for @NousResearch’s Hermes Agent. Our team kept coming back to Hermes for the same reason: it is stable, and it shows real self-improving behavior over time. So we added first-class support for it in ClawShell. This release adds: • Hermes Agent support • MiniMax as a provider option • /admin/stats with persistent runtime counters • onboard-installed stats skill + weekly stats cron • Bug fixes

ClawShell v0.1.1 is out. We added email boundary protection. Reason: one of our engineers asked what happens when an agent can read banking OTP emails. Fair question. This release adds: •Email credential isolation •Allow / block lists for mailbox access •OpenRouter support •Bug fixes If you’re running agents with real credentials, email is part of your attack surface whether you like it or not. This is one step toward shrinking that surface.

@ArbRoyalsTheAce If agent can access your keychain directly, they can be exploit easily. Agent execution must be treated as adversarial compute

@runtainc We have placed all of our keys onto the Apple keychain and do not leave them in the env . Have you seen any exploits

We have been exploring this area, and we have released ClawShell to isolate the 2 & 3 : basically Linux permissions for AI agent secrets. Instead of trying to prompt-filter your way to security, just isolate sensitive ops at the process level. Details here: x.com/runtainc/statu… Code: github.com/clawshell/claw…

The power of AI agents comes from: 1. intelligence of the underlying model 2. how much access you give it to all your data 3. how much freedom & power you give it to act on your behalf I think for 2 & 3, security is the biggest problem. And very soon, if not already, security will become THE bottleneck for effectiveness and usefulness of AI agents as a whole (1-3), since intelligence is still rapidly scaling and is no-longer an obvious bottleneck for many use-cases. The more data & control you give to the AI agent: (A) the more it can help you AND (B) the more it can hurt you. A lot of tech-savvy folks are in yolo mode right now and optimizing for the former (A - usefulness) over the the latter (B - pain of cyber attacks, leaked data, etc). I think solving the AI agent security problem is the big blocker for broad adoption. And of course, this is a specific near-term instance of the broader AI safety problem. All that said, this is a super exciting time to be alive for developers. I constantly have agent loops running on programming & non-programming tasks. I'm actively using Claude Code, Codex, Cursor, and very carefully experimenting with OpenClaw. The only down-side is lack of sleep, and an anxious feeling that everyone feels of always being behind of latest state-of-the-art. But other than that, I'm walking around with a big smile on my face, loving life 🔥❤️ PS: By the way, if your intuition about any of the above is different, please lay out your thoughts on it. And if there are cool projects/approaches I should check out, let me know. I'm in full explore/experiment mode.