firecracker is an obvious place to start because it offers best-in-class support fast cold starts and hard isolation. we can't speak to what exactly the other browser infra providers are doing, but based on public benchmarks we're up to 5.8x faster
As part of launching our MCP server, we wrote a short blog post about some of our technical decisions and limitations of the current landscape. The space still feels very early, so would love others' thoughts on the MCP space and what else is needed / how you think it'll evolve over time.
The big motivation we had for trying to do this on a unikernel is actually the session pause/resume. The fast cold restart is a nice addition (or put the other way, a super slow restart would prohibit session pause/resume from being useful)
Yeah, this is mostly just an OSS implementation for self-hosting! I'm not entirely sure what any of those companies use under the hood, in theory they could use this implementation if they wanted to gain the fast boot times, session reuse etc.
Haven't tried on a Lambda. Would be curious to know the results if you do!
It took us 8gb to get it up - maybe could be slimmed down if you took out the Anthropic Computer Use components, but browsers generally are pretty heavy.
Thank you! Yeah, the current implementation basically performs the same as a Docker container (i.e. not much). The interface is the same, so you can use BU/Playwright/Puppeteer's header configs to change as needed.
We did notice the unikernel cloud instances don't run into bot detection as often as our hosted docker instances, but I think that's mostly because Cloudflare haven't flagged Unikraft Cloud's IPs yet, hah.
Your demo is awesome, nice job. Do you guarantee a certain response time?
Also, I'm curious how people use your API if it takes 30 seconds+ to get a response -- ie can I use this for time-sensitive annotations, and also, wouldn't my code hang waiting for your response or do you offer webhooks?
Super good point that we've found with our customers who are migrating from outsourced bookkeepers.
We're using our Slackbot as a friendly way for companies to get their financial data, the idea being that if the numbers are accessible/convenient and ones they want to know (revenue, burn rate, bank balances). Then, we can use the same Slackbot to provide other necessary financial details.
We're still working on this, but we think Gusto does a great job of communicating necessary (and not so exciting) info in a way that engages us.
Xero and Quickbooks just use basic regex to match transactions, which doesn't make much sense if the transaction category isn't static, eg Amazon. We also learn categories on an aggregate level, so if your bank statement's never seen a transaction from, say Tommy's Tacos, but someone else has, we'll be able to categorize it correctly immediately.
For categorization, we're using ML (which learns over time and isn't rule based like Quickbooks' suggestions, which uses regex).
For the other pieces of bookkeeping, we're using integrations to automatically match invoices/payments (especially disjointed ones, eg invoice through QBO but pay separate via Stripe on client's website). We're also doing stuff like adjusting journal entries programmatically, like payroll accruals. For now, QB is providing the financial reports, we're doing the bookkeeping so their automated reports are accurate!
We're basically doing the services that a bookkeeper would provide on top of a client's Quickbooks software. QB is a tool and we aim to be a service.