1
So we built a runtime to make autonomous use safer. Railyard is an open-source runtime that sits between Claude Code and the shell and adds guardrails to agent commands.
Every command Claude runs goes through Railyard first. Most commands pass straight through. The ones that could cause damage (for example terraform destroy) get blocked or require approval.
Under the hood it runs commands inside an OS-level sandbox (sandbox-exec on macOS and bwrap on Linux) and applies deterministic rules. There’s no LLM scoring commands or guessing about intent — a command either matches a rule or it doesn’t. The check takes about 2ms.
By default it blocks destructive commands like terraform destroy or rm -rf, prevents access to sensitive paths like ~/.ssh, ~/.aws, and /etc, restricts certain network calls, and catches simple evasion tricks like base64, hex, or variable obfuscation.
It also snapshots file writes so you can roll back a session if something goes wrong.
In practice this lets us run Claude Code with --dangerously-skip-permissions, but with guardrails underneath so we can move fast without breaking or deleting production assets.
We built this because we wanted Claude Code to behave more like a software factory. Factories run at high volume, but only because the production line has quality and safety checks. Railyard is the guardrail layer that makes that possible for us.
Repo: https://github.com/railyarddev/railyard
It's MIT licensed and free to use. If you're experimenting with autonomous agents, feel free to clone it and try it out. I'm especially curious how people push or break these guardrails.
Happy to answer any questions about how it works.
I would love some feedback on our product:
1. How is the landing page / website
2. What use cases could this unlock for you?
We have some awesome tech behind the scenes and have some initial customers, but I am really keen on expanding towards a PLG motion where folks can sign up and get insights about their data fast.
Would love brutally honest feedback! See it here: https://arka.so
I would love some feedback on our product:
How is the landing page / website
What use cases could this unlock for you?
We have some awesome tech behind the scenes and have some initial customers, but I am really keen on expanding towards a PLG motion where folks can sign up and get insights about their data fast.
Would love brutally honest feedback!
I think we’ve essentially unlocked unlimited engineering capacity for routine tasks, thanks to these AI tools.
First, we built everything serverless and chose Vercel. The bigger cloud vendors are great, but unless you have hundreds of customers, they’re often overkill. With serverless, unless there’s usage, there’s no cost. At some point the cost will exceed the value, but we’re not there yet. Vercel is serverless-first. This is not an ad, our usage has not crossed $20/month (compared to thousands somewhere else).
Second, DuckDB has become our in-memory engine for analytics where, in the past, we would’ve needed to run something like Snowflake. Even for medium analytical tasks (parse and analyze a massive CSV), DuckDB is brilliant. We’re also experimenting with making this scale by building our own DuckDB cluster and trying tools like MotherDuck to see how the cost compares.
Third, we heavily use Claude Code Web. Once Claude familiar with your repo (ask it to make copious notes), async coding is the next level of productivity. I now ask it to write a lot of the repetitive code paths. I put my kid to sleep, kick off 10 tasks, and wake up with usable code the next morning. I’m sure Codex is good too.
Fourth, we made everything testable and runnable by LLMs. Every new piece of functionality comes with clear debug routes that an LLM can understand and iterate on. We tell the LLM: you build it, you test it, you fix it, you Q/A it, come back with a report.
We also built everything in containers, so we’re not locked into a single cloud vendor and can deploy into a customer’s own environment easily. LLMs also understand these services the best, because they’re open-source.
So what does this mean?
There’s a lot of talk about human vs AI productivity right now, but I’m actually super bullish about the future. I think we’re going to see many more David-vs-Goliath stories. Small teams will win more often. We will have more time to build things the right way, and more time to talk to customers and learn.
Sharing this in case it’s useful to others building early.
The motivation is personal: at every company I’ve worked at, our data was scattered across warehouses, Slack, GitHub, Asana, and a dozen other tools. I spent way too much time copy-pasting from one source to another just to answer basic questions.
This project is still really early and rough around the edges, but it can already generate dashboards, summaries, and insights without any modeling or setup.
I’m genuinely not sure how useful this is beyond my own workflows, so I’d really love some brutally honest feedback from y'all.
Here’s a short video of it in action if you’re curious: https://www.youtube.com/watch?v=HTp8flEeZao
And you can try it here: app.arka.so
I'm Ari, I previously led AI at Asana and built databases at AWS.
We built Arka for startups who need insights from their data but don’t have the time (or team) to build out a full analytics stack.
Does this sound like you?
1. Your investors, customers, product, marketing, and ops are all asking for answers 2. Your data lives across Postgres, Stripe, Segment, maybe Databricks, even Slack 3. But your dashboards are out of date, writing the right query takes forever, and reporting is a mess
Arka is your entire data analytics stack in a box:
1. Connects to your databases and SaaS tools instantly 2. Auto-generates clean metrics from messy data 3. Lets anyone ask questions in plain English — no SQL or Python needed 4. Builds reports, schedules alerts, and automates insights
Think: Metabase + dbt + automated reporting, without staffing a full data team.
It’s still early, but Arka is already helping early adopters get up and running with analytics faster than anything they’ve used before.
We’re now onboarding a few more early adopters. If you're a startup, we’ll personally help you get set up: from connecting your tools to defining clean metrics and building out reports.
No cost during early access. Just bring your data, we’ll handle the rest.
Try it here: https://arka.so
Or drop a comment, super happy to jam your data pains!
I’m excited to introduce Mode, now in early access! Mode is a Visual Studio Code extension I’ve been working on, acting as your personal AI code copilot. It’s a reliable backup to coding tools like Cursor, especially when you run out of free tokens.
Try it out here: https://marketplace.visualstudio.com/items?itemName=aruna-la...
Why Use Mode:
Bypass Token Limitations: If you find yourself copy-pasting code into ChatGPT or Claude after running out of free tokens, Mode is for you. Mode lets you to use your own API keys and subscriptions, giving you uninterrupted, seamless AI-driven coding assistance.
Chat and Merge Features: Mode currently focuses on chat-based assistance with a Cursor-like merge feature. Autocomplete is coming soon!
Direct Model Upgrades: With Mode, you can directly upgrade to the latest LLMs (e.g., o1-preview) without any additional costs often associated with other tools.
No Backend Overhead: Mode is essentially a smart client, and doesn’t have its own backend. It connects directly to providers like OpenAI, meaning you keep full control over your API keys, data, and preferences.
Why I Built Mode:
I love AI coding tools like Cursor, but token caps, throttling, and downtime make serious coding frustrating and/or expensive.
Try Mode:
Mode is in early access, and your feedback would be super helpful in shaping its next steps. I’m prioritizing building features that matter most to y’all. I’d love for you to give Mode a try. Looking forward to hearing your thoughts and suggestions!