Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- If AI has achieved anything it's a dramatic increase in documentation quality for the sake of training or instructions.
I'd argue that if documentation on how to operate your business can be used to feed an agent, your problem was never the ability to run operations and the ROI will amount to the ability to ask instead of doing. I don't know how much this is gonna go for, but you could probably have one of your existing staff just do basic documentation lookups too.
by pluc - The hard part for an always-on company agent is what survives a restart. My runtime snapshots the whole JS heap to bytes and restores it in a fresh process, so conversation and working state come back with no serialization code — but timers don't survive, so an agent re-arms them from declarative state after restore. How do you handle that? Is a long-lived agent's state checkpointed, or rebuilt by replaying context on each wake?
- Don’t have time to dig deep into it this afternoon, but want to comment to come back but also to suggest an expansion past your stated “one personal and one company” wiki.
I lead two different functional areas within tech and work in a larger tech platform for the parent company of a couple dozen operating companies.
All that to say “I would almost surely be better served by a hierarchy or general graph of wikis rather than just ‘me’ or ‘all 15K employees’.”
Even a 50 person company might benefit from “me”, “my team”, “my department”, and “whole company”.
by sokoloff - The website tells me absolutely nothing about what makes this new or uniqueby dbbk
- Congrats on the launch, the approach of spending compute upfront is a compelling shift from what the current standard is.
I am curious as to how you handle state invalidation and unseen side-channels during those long-running tasks, because if the agent pauses a task for X days, say waiting for an email reply, but the projects params changed offline or in a channel that the agent has no eyes on, how does the system realize that the pre-compiled context has now gone stale? Does that also mean that proactive actions always require a HITL?
by i18ner - Congratulations on the launch! The demo in the video you linked looks awesome (though your video resolution is a bit low, some of the text is hard to read on screen, I suggest you record in higher resolution). I think you made a really good choice to have your connector provider handle OAuth credentials, since it is really hard to get security right on a multitenant app like yours, where some of your agents are doing work on customers' behalf and others are improving your product itself :). The design of your app interface also looks really great!by logicallee
- 7-day trial seems kind of narrow for a product that needs to "build a memory", no?
- There is a lot of competition in this space, both commercial companies and open source. It seems like the biggest advantage for using one of these and not Claude or Codex is that you can use multiple models. Otherwise, those two have all the same features, or probably will in the next week.
Are you able to use multiple models? Or I guess the first question is, what model(s) are you using? At this point, many enterprises care so that should be front and center, especially if you're using models hosted in China, because a lot of non-Chinese companies care about that.
You mention some things about unique ways to store memory. Do you have any data or use cases on how that improves performance? If you do, you should get those on your website too.
I'm not sure you have a very strong answer to "Why not just run Claude or Codex on my laptop?" Sure, you provide compute, but so do they, and most enterprises and even regular people are getting desktops/droplets to run their agents now because of this problem. This is kind of solved already.
Also there is a small nit with your website. The graphic next to "Proactively gets things done for you" gets bigger and smaller, so if you're trying to read anything below that, it keeps jumping up and down.
by jedberg