AgentSky is a cloud runtime that lets developers call Claude Code, Codex, Hermes, OpenClaw, pi, DeepSeek Harness, Kimi Code, and opencode through a single API, positioning itself as an OpenRouter equivalent for coding agents rather than raw models. The platform pairs each agent with a model, connects it to channels like Slack, Telegram, or a custom API, and keeps persistent sessions so state carries between messages, with pricing structured as per-minute agent time plus per-token model usage. The pitch to teams is that agents which normally run on a developer's laptop can be cloned into the cloud with one command, keeping the same instructions, model, and MCP servers.
The launch matters now because agent quality and cost vary wildly across harnesses, and teams are actively swapping between them. The founder points out in the launch tweet that the same task ran for $150 on one agent and $2 on another, and the accompanying Agent Arena lets users benchmark harnesses head to head on real GitHub tasks by quality, speed, and cost. For users who already pay for Claude Pro, Claude Max, or ChatGPT, connecting the subscription once lets eligible agents run without a second model bill.
AgentSky is built by Xiaoyin Qu (@quxiaoyin), previously a Meta PM and founder of Run The World (backed by a16z and later acquired), and more recently the founder of heyBoss.ai. The product is aimed at developers running one agent for themselves and at companies running thousands in parallel for what the team calls a software factory, where an orchestration layer routes tasks across Claude Code, Codex, and DeepSeek-style harnesses without rewriting the backend each time a better agent ships.
Comments (15)
Sign in to join the discussion.
N
Nadia BurkovAug 24, 2026
The $150 vs $2 hook is doing serious lifting in that tweet. Whoever wrote that copy earned their coffee this week.
J
Juliette VoAug 24, 2026
The pricing chart in your docs is doing more marketing than the tweet did. Put that screenshot in the pinned post.
P
Pablo NtembeAug 24, 2026
Rate limits per agent or pooled across the account? And do you normalize the streaming format or do I still have to write six parsers?
K
Kira OkaforAug 24, 2026
OpenRouter for Agents is a very tidy positioning line. My only note is that Claude Code alone isn't really a comparable to a $2 harness, so the framing invites nerds to nitpick.
T
Toshi Y.Aug 24, 2026
Tagline rewrite, on the house: 'Every agent. One key. Pick your price.' You can Venmo me later.
B
Beatriz SalgadoAug 24, 2026
Curious about the take rate here. If you're passing through provider costs plus a cut, the margin math gets spicy fast when someone burns a Claude Code loop overnight.
A
arjunbuildsAug 24, 2026
How many people shipped this? Asking because I've been staring at my own agent router repo for two months and now I feel personally attacked.
R
Ravi K.Aug 24, 2026
I was literally three weeks from launching this exact idea. Congratulations, I now have to pivot to 'OpenRouter for OpenRouters.'
L
Linh PhamAug 24, 2026
Been watching this team for a bit. The bet that agents commoditize before models do is the correct one, and this is the pickaxe.
M
Max VoelkerAug 24, 2026
Every provider is going to ship their own hosted harness within a quarter. Aggregators in this space have a shelf life measured in vibes.
S
Sana IqbalAug 24, 2026
No launch video? Feels like a missed layup for a product where the demo is literally 'watch six agents fight and one wins for $2.'
C
Chidi OkonkwoAug 24, 2026
Signed up. First tap I get thrown into a docs page with six SDK snippets and no 'try it now' box. Give me a playground before you give me a curl command.
R
roshi.exeAug 24, 2026
The agent you can't afford to run is the one you learn the most from.
M
Mimi JanssonAug 24, 2026
Recruiting brain kicking in: who's picking up infra hires? I have an SRE who has been benchmarking harnesses for fun on weekends.
S
seb.ethAug 24, 2026
Waiting for the moment where each agent call settles onchain and we can arbitrage Claude vs DeepSeek pricing in a dark pool. You know it's coming.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
AgentSky
Run Claude Code, Codex, Hermes, OpenClaw, Pi, and more in the cloud through one API.
Run Claude Code, Codex, Hermes, OpenClaw, Pi, and more in the cloud through one API.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
AgentSky is a cloud runtime that lets developers call Claude Code, Codex, Hermes, OpenClaw, pi, DeepSeek Harness, Kimi Code, and opencode through a single API, positioning itself as an OpenRouter equivalent for coding agents rather than raw models. The platform pairs each agent with a model, connects it to channels like Slack, Telegram, or a custom API, and keeps persistent sessions so state carries between messages, with pricing structured as per-minute agent time plus per-token model usage. The pitch to teams is that agents which normally run on a developer's laptop can be cloned into the cloud with one command, keeping the same instructions, model, and MCP servers. The launch matters now because agent quality and cost vary wildly across harnesses, and teams are actively swapping between them. The founder points out in the launch tweet that the same task ran for $150 on one agent and $2 on another, and the accompanying Agent Arena lets users benchmark harnesses head to head on real GitHub tasks by quality, speed, and cost. For users who already pay for Claude Pro, Claude Max, or ChatGPT, connecting the subscription once lets eligible agents run without a second model bill. AgentSky is built by Xiaoyin Qu (@quxiaoyin), previously a Meta PM and founder of Run The World (backed by a16z and later acquired), and more recently the founder of heyBoss.ai. The product is aimed at developers running one agent for themselves and at companies running thousands in parallel for what the team calls a software factory, where an orchestration layer routes tasks across Claude Code, Codex, and DeepSeek-style harnesses without rewriting the backend each time a better agent ships.
Comments (15)
The $150 vs $2 hook is doing serious lifting in that tweet. Whoever wrote that copy earned their coffee this week.
The pricing chart in your docs is doing more marketing than the tweet did. Put that screenshot in the pinned post.
Rate limits per agent or pooled across the account? And do you normalize the streaming format or do I still have to write six parsers?
OpenRouter for Agents is a very tidy positioning line. My only note is that Claude Code alone isn't really a comparable to a $2 harness, so the framing invites nerds to nitpick.
Tagline rewrite, on the house: 'Every agent. One key. Pick your price.' You can Venmo me later.
Curious about the take rate here. If you're passing through provider costs plus a cut, the margin math gets spicy fast when someone burns a Claude Code loop overnight.
How many people shipped this? Asking because I've been staring at my own agent router repo for two months and now I feel personally attacked.
I was literally three weeks from launching this exact idea. Congratulations, I now have to pivot to 'OpenRouter for OpenRouters.'
Been watching this team for a bit. The bet that agents commoditize before models do is the correct one, and this is the pickaxe.
Every provider is going to ship their own hosted harness within a quarter. Aggregators in this space have a shelf life measured in vibes.
No launch video? Feels like a missed layup for a product where the demo is literally 'watch six agents fight and one wins for $2.'
Signed up. First tap I get thrown into a docs page with six SDK snippets and no 'try it now' box. Give me a playground before you give me a curl command.
The agent you can't afford to run is the one you learn the most from.
Recruiting brain kicking in: who's picking up infra hires? I have an SRE who has been benchmarking harnesses for fun on weekends.
Waiting for the moment where each agent call settles onchain and we can arbitrage Claude vs DeepSeek pricing in a dark pool. You know it's coming.