The Context and Control Model for Production
Joined June 2024
- Tweets156
- Following497
- Followers89
- Likes127
NOFire AI retweeted
I'd rather put a coding agent in its own microVM than keep auto-approving it on my host OS, especially when open weights this size are actually downloadable.
Naive-N0.5-Flash. MIT 309B MoE with 15.5B active and a native 1M context, built with no full-attention layers. Weights are up today; announced API pricing is $0.10 / $0.40 per million tokens, though the API is not live yet.
OpenScience. Free Apache science workbench that reads papers, writes and runs code, and Autoresearch hill-climbs experiments on Modal, your machine, or Slurm. Install with npm and open a browser workspace.
Brig. Go CLI plus daemon that runs agents like Claude Code inside a microVM on Apple Silicon or Linux, so the agent gets its own kernel. Apache 2.0; landed on Show HN yesterday.
Push back if you'd still auto-approve on the host and treat the open MoE as a curiosity.
NOFire AI retweeted
Tired of coding agents that can touch your whole laptop from one bad prompt?
Try brig (brig-sh/brig) - a Go CLI that runs agents inside a microVM on your Mac or Linux box. One project mount, no host keychain by default, trash the sandbox when done.
NOFire AI retweeted
Bookmark this if you run Claude Code / Codex with auto-approve.
Your agent can rm, curl, and leak keys. Sandbox map:
Local microVM (real VM on your machine)
• Docker Sandboxes - sbx run claude
• shuru - ephemeral, offline by default
• Brood Box - microVM + review diffs before they land
• Brig - open-source microVM, curated agent profiles
# curl -fsSL brig.sh/install | sh
Cloud (agent runs elsewhere)
• E2B - Firecracker, SDK-first (most used for products)
• Daytona - persistent agent workspaces
• Vercel Sandbox / Modal - hosted code exec
Weaker (shared kernel)
• DevContainers / plain Docker - fine for deps, not a blast-radius wall
Pick by job:
laptop agent → Docker SBX / shuru / Brig
Product feature → E2B
NOFire AI retweeted
Jev (RLCD) opens exciting opportunities. A general purpose classifier is what many usecases are after. LLMs are the wrong or expensive choice for many of the problems they are used for today.
Top open Qs? Training data set, what is it? shape/origin? How's the model eval'ed?
NOFire AI retweeted
While working on Ravel we noticed arrayef getting yanked, see: github.com/NOFireAI/ravel/co…, decided to keep the 0.3.9 pin. Now assessing.
This is why we develop in our #sandbox solution.
We are soon announcing the FOSS release of our sandboxing solution to the public.
Stay tuned.
NOFire AI retweeted
Hugging Face could reconstruct 17,000+ attacker actions because they own the infrastructure the agent ran on, end to end. Most teams running agents today don't have that. linkedin.com/pulse/agent-wro…
NOFire AI retweeted
One of the reasons why proper sandboxing and hard controls matter. That’s what we’ve been building at @NOFireAI… More details soon :)
NOFire AI retweeted
At @NOFireAI we 'design for breach'. The AI agent in production is untrusted by default. nofire.ai/blog/design-for-br…
NOFire AI retweeted
Nine years at @elastic , shipping the #observability products other teams got paged on. Watched the same pattern every time: AI tools arrive, talk big, then ask the on-call engineer to retrieve more context.
Today we announce what we built instead..
NOFire AI retweeted
Remove the human, ship faster, no approval needed.
Then the CISO walks in and the room goes quiet.
@Gartner_inc predicts 40% of agentic AI projects will be canceled by end of 2027.
#AISRE #Kubernetes #ReliabilityEngineering #SRE
NOFire AI retweeted
Yann LeCun at GTC 2025: "Scaling text prediction won't get us to real intelligence." This matters for production engineering. When your system fails, you don't need LLMs generating plausible explanations.
#SRE #CausalAI #OnCall #ProductionEngineering #ReliabilityEngineering
NOFire AI retweeted
Just like in #DevOps, success in life is about collaboration over competition. Foster a culture of trust, embrace failures as lessons, and automate the mundane. Remember, a resilient team is your best infrastructure! 🚀✨ #Leadership #TechCulture #PlatformEngineering #DevSecOps
NOFire AI retweeted
We talked about alert chaos and knowledge silos.
The root cause? We think about reliability AFTER we build.
Write code → Deploy → Fix when it breaks.
The best engineers ask "how will this fail?" while they're coding.
Reactive vs proactive thinking.
NOFire AI retweeted
"When you have 100s of alerts firing daily, doing RCA on everything is impossible. We need to focus on fixing the foundation first."
This captures the gap between vendor promises and engineering reality.
#ReliabilityEngineering #Kubernetes #SRE
NOFire AI retweeted
During your last incident, how many tools did you check?
Prometheus → Datadog → Grafana → ELK → PagerDuty → Slack
Each with different query languages, different data models, different ways of representing the same problem.
#SRE #OnCall #ReliabilityEngineering #Kubernetes