What Capy is, and how Morph Ultra fits in

Capy comes from Scrapybara, a San Francisco company. It’s a cloud product for running fleets of coding agents. On 11 November, Capy and Morph said you can plug a Morph Ultra API key into it and run those agents on Kimi K3.1 and DeepSeek V4.1 Flash.

Earlier this week I covered Morph Ultra itself. It’s $200 a month, and Morph is selling only 1,000 subscriptions. There’s no token meter on Kimi K3.1 or DeepSeek V4.1 Flash, but there are per-minute limits, and Morph can change them. I want to set out what Capy is, what it costs, and how the two bills work once you connect them.

Unlimited Kimi K3.1 and DeepSeek V4.1 Flash, now in Capy.

Connect your @morphllm Ultra key and run Capy agents on both models for one flat $200/month.

— Capy (@capydotai) November 11, 2026

What Capy is

Capy describes itself as a development environment for a world with many agents at once, not a text editor with autocomplete bolted on. Work is organised around tasks — issues, features, bugs — rather than open files. You start threads from Slack, Linear or the web, and Capy opens pull requests on GitHub.

The company behind it is Scrapybara, Inc., a Y Combinator Fall 2024 startup in San Francisco. Nalin Semwal is CEO and Justin Sun is CTO. In Capy’s model, Captains are long-lived orchestrators, and each cloud thread runs on its own isolated Linux Firecracker microVM in US East (New York), on Ubuntu 24.04. Capy says it is SOC 2 Type II, with encryption in transit and at rest.

You can run models on Capy’s own credits at the provider’s list price, or connect your own subscriptions and API keys. Capy does not charge credits for the tokens those keys serve. Machine time, generated images and reviews still come off the organisation’s Capy credit balance.

Morph Ultra inside Capy

Connect a Morph Ultra key, pick Kimi K3.1 or DeepSeek V4.1 Flash, and Capy agents run on those models for Morph’s flat $200 a month. Capy charges no credits for the tokens the Morph key serves, on every Capy plan. A thread on a cloud machine still uses credits for machine time. A thread on your own computer doesn’t.

The steps, from Capy’s post and Morph’s product page, are: subscribe to Morph Ultra and create an API key; in Capy, open Settings → Models, add a connection, choose Morph Ultra, and paste the key; then pick one of the two included models when you start a thread. A key badge on the thread means it’s running on the Ultra key. Morph’s Capy page is clear that Ultra is a plan for one person, so the key goes on your own connections, not the organisation’s.

Capy’s 0.8.0 changelog already listed the Morph Ultra connection on 10 November. The blog post and the posts on X went up the next day.

For Morph’s Nick Khami, the point of the partnership is cache hits. He said cache hits across Ultra subscribers are 93%, and that a harness like Capy can get much closer to 99%, which is where he says the subscription’s value shows up. Morph Ultra’s docs say cached input does not count toward the plan’s per-minute limits.

excited to announce this partnership with @capydotai

morph ultra offers unlimited cached tokens. right now cache hits across all subscribers are only 93% which is less than ideal.

optimized harnesses like capy let you hit much closer to 99% which is where the value of the sub really shines.

— Nick Khami (@skeptrune) November 11, 2026

Two bills, not one

This is not a bundled product. You pay Morph for Ultra, and you pay Capy for Capy.

Morph Ultra is $200 a month, limited to 1,000 subscriptions. There is no token meter and no monthly token cap on the included models, but unlimited does not mean unthrottled. Each subscription has per-minute limits on uncached input and output, plus a burst reserve, and requests can return HTTP 429 when Morph is at capacity. Morph can change those limits, and the plan has no uptime SLA. I went through the plan, the limits, and what happened when I used it in Amp Code, in the earlier piece.

Model	Uncached input per minute	Output per minute	Uncached input burst	Output burst
Kimi K3.1	24,290	6,162	72,870	18,486
DeepSeek V4.1 Flash, both versions	400,000	70,000	1,200,000	210,000

Other models on Morph still bill per token. The plan includes no credits.

Capy’s own plans set monthly credits and how much cloud machine capacity you can run, with unlimited members and no seat fees:

Plan	Monthly	Annual	Included credits
Lite	$20	$16 a month	$20
Pro	$100	$80 a month	$105
Max	$200–$1,000, in $100 steps	$160–$800 a month	110% of the monthly price
Enterprise	Custom	Custom	Custom

Unused credits don’t roll over. You can add balance at 1:1, with a $5 minimum.

Cloud machines bill only while they’re awake, plus a two-minute wind-down each time they sleep. A sleeping machine costs nothing. The default Large machine is $0.40 an hour; sizes run from Small at $0.10 an hour up to Big Guy at $3.20 an hour, with larger sizes locked to higher plans. Machine time is always paid in Capy credits, even when the model runs on your own key.

Capy also has a machine size named Ultra — 8 vCPU, 32 GB, $0.80 an hour, on Pro and above. That’s a VM size, not the Morph plan.

If you want to subscribe to Morph Ultra, I have a referral link to Morph’s sign-up page. It goes to Morph’s ordinary sign-up, and I may earn a commission if you use it.

Who it’s for

Capy is for people who want to direct several agents at once from Slack, Linear, GitHub and the web, and who are happy to pay for isolated cloud machines, or to run a thread on their own computer. Connecting Morph Ultra is for people who already have that $200 plan, or who want those two models without a token meter, and who’d rather Capy not bill tokens for them.

You don’t need Capy to use Morph Ultra. The plan works in other coding agents, including Amp Code, which is how I’ve been using it. And you don’t need Morph Ultra to use Capy: you can run Capy on its own credits, or on other keys you already pay for.