The Claude Code background agent you needed. Deploy it in our cloud or yours. Native agent @ mentions in Linear, Slack, Github.
Joined August 2025
- Tweets53
- Following47
- Followers215
- Likes135
Cyrus retweeted
@CyrusAgent offers a free and open source way to plugin to that agent platform on Linear, and connect Claude Code or any harness, and build YOUR inspect that you own
A pattern I keep seeing at the most AI-forward companies: they stop shopping for an agent and start building one.
@zachbruggeman at @tryramp built Inspect, a coding agent that knows Ramp's codebase, conventions, and test suite. You hand it a Linear issue the same way you'd hand it to a teammate. It reads the docs, writes the code, runs the tests, and comes back when they pass.
Off-the-shelf agents are trained on the world. Yours is trained on your company, which is why we've built Linear to be the place any agent plugs into. Your agent, your rules, one system of record for everyone doing the work.
Cyrus retweeted
Cool investigative work from @entrecurious's team to benchmark the DGX station for local and open inference! @NVIDIAAI's machine is slowly rolling out, and fun to see what people do with it.
Recently at Cornell I used it for synthetic RL environments and agent training---high risk moonshot research that turned out to be high reward! kilian-group.github.io/blog/…
I squeezed lot of compute out of the DGX station for RL training. The bottleneck was throughput on agent rollouts, so the machine's optimizations for agent inference actually enabled fast training.
Can you run the frontier at home, at work?
We dug deeper to find out every detail of the DGX Station, and what kinda hardware it's taking to run GLM 5.2.
Is the GLM they say is running the same GLM as the benchmarks? Not always, but that doesn't mean it's a write-off.
atcyrus.com/stories/dgx-stat…
Cyrus retweeted
50 Demos in 50 Days, Day 3 Demo.
Running Kimi K2.7 using openrouter and opencode INSIDE @linear using @CyrusAgent.
I found out EXACTLY how many times cheaper it is to run K2.7 Code and if it really performs like Opus.
That's Claude Opus 4.8 vs Kimi K2.7. And in our testing the cheaper one made a design leap we didn't expect.
Full run inside Cyrus 👇 [1 min watch]
Cyrus retweeted
Demo 2/50!
ScheduleWakeup tool!
This is incredible tool you NEED to be making use of if you are not already, and thankfully, we've now fixed a bug with it in Cyrus.
Cyrus retweeted
I'll be doing Cyrus office hours, an open invitation to come and chat Cyrus, Claude Code and Linear.
Cyrus is the open framework to build your company's internal agent.
luma.com/j906tnz9
Cyrus retweeted
just days since @cursor_ai announced the cursor sdk, and WHILE at stripe sessions we shipped improved cursor support for @CyrusAgent with it.
Let composer-2 cook!
thanks especially to @SamSokolin who quickly triaged sandboxing which was broken in the first release on our feedback.
Cyrus retweeted
will be switching to it from the bare cursor cli for @CyrusAgent cursor support! sweet!
Cyrus retweeted
@CyrusAgent stars are growing rapidly.
It's because it's the only open source agent that gives you a
- @linear agent
- @github agent
AND
- @SlackHQ agent
In one installation, to tackle your end-to-end software automation
Cyrus retweeted
Today, on Launch Week Day 2, we are launching something that's been requested many times, and for good reason.
Directly @ mention Cyrus on Github (@cyrusagent) to get a response!
You can get Cyrus to leave a code review, to answer quick questions about the PR, or to submit a follow up Linear issue for something you see.
AND Cyrus now gets notified if you leave a review on a PR opened by Cyrus with a "change request", offering the chance to automatically get those changes in.
And we're not done yet, tune in tomorrow for one more major launch to get excited about. Cyrus will be joining the chat.
Cyrus retweeted
Exciting customer news 🚀
@mihail_eric the @MonacoGTM team are out of stealth.
They’re building the first AI-native revenue engine for startups—automating everything from ICP discovery to personalized outreach to CRM workflows.
The incredible team is backed by @foundersfund, @garrytan, and @peterthiel.
Monaco uses @CyrusAgent to ship faster. When we first spoke, Mihail shared how context switching was slowing his team down. Watching him solve that—and now launch something that will help thousands of founders—is what startups are all about.
Grateful to be part of Monaco’s journey. Can’t wait to see what they ship next with Cyrus!
Cyrus retweeted
For product builders, this new world is INSANE.
@meetgranola → @claudeai → @linear → @CyrusAgent
Granola records all-hands. Claude transforms into actionables, files in Linear with owners.
Humans tackle priority, user-facing items with Claude. Cyrus tackles minor issues.
Cyrus retweeted
Shipping velocity is fast and furious @CyrusAgent and now we have the receipts:
atcyrus.com/changelog
If you're still sleeping on the chance to add Claude Code to your @linear workspace you're missing out on major alpha for long-running agents.
Cyrus retweeted
BREAKING: Claude Code is soon launching Teammate/Swarms 🤖🤖🤖 and I have the receipts.
They'll be bringing rich agent swarm functionality surely before the year end, as it's almost all baked in to the CLI and agent sdk now! Issue tracking, mailboxes, read/write locks.
FULL analysis check it out for yourself!
git.new/AYVq752
Cyrus retweeted
The Toronto Claude Code Meetup was epic fun! Especially demo-ing @CyrusAgent and how we overcame the issue tracker - PR not creating trash PRs problem.
Thanks @claudeai @surim0n and @robjama for having me out. Trip from Victoria was 100 worth it to meet the brilliant minds here!
Cyrus retweeted
From complexity to simplicity in ten seconds.
Struggling to make changes to that complex part of the codebase?
Ask Cyrus to map out the state and flow in mermaid and post it to @linear as a comment!
Instant understanding.
Cyrus retweeted
8 months ago, I hit a wall.
Pressure was sky-high. As a technical co-founder, I was the sole person who could tackle our outdated dependencies. As a small and early stage team, hiring was off the table. Every weekend blurred into the next as I threw every AI tool at the upgrade task, yet nothing worked.
We were stuck.
I couldn’t let it go. For months, I obsessed over how to get AI to handle messy, high-stakes engineering work-not just toy problems. Most models just fizzled out when faced with the tough stuff.
Fast forward to last weekend: I ran the same upgrade task again. This time, our own @CyrusAgent powered through 6 straight hours of breaking changes, end-to-end. I came back on Monday, and it was done.
Now that we're here, I feel like I have my creative time back and I'm not stuck on the codebase "chores". There's so much that's possible.
Welcome to Launch Week - Day 1 for @CyrusAgent
Cyrus retweeted
I can now make code bases with @CyrusAgent of concepts I’ve carried around for years, that I’ve never had the time, talent, or capital to execute, faster than I can read their documentation.
Design + dev, at the speed of thought. This is good.
Cyrus retweeted
3 hours deep in flowcraft with @CyrusAgent. So far this session we’ve committed 6 PRs on Forgetmenot, built a watchOS app, and now working through phase 2 of voice command integration. Months of planned work, and I just finished my coffee..