Launch HN: Grapevine (YC S19) – A company GPT that actually works (getgrapevine.ai)
Anyone can setup a Grapevine Slack bot and have it respond to and optionally proactively answer questions that require company context. Here’s a demo video with examples: https://youtu.be/_nrfbZzvxU8
We built Grapevine because we were interested in a ChatGPT that fully understands your company. We'd tried many of the existing tools (including expensive enterprise ones), but while they were good at answering “what is X team’s Q4 goal,” they weren’t good at the day-to-day questions that actually blocked people.
So, our founders and early engineers created a set of 100+ representative questions, from hard technical questions to company-specific knowledge questions. At first, the state-of-the-art “enterprise search” products were getting about 50% of them correct, and our in-house system was getting 35%. But as we solved details in data processing, search algorithm, and more, we eventually achieved 85%. (For reference, the best human score our founders got was 70%)
It’s changed the way we work: popular engineering channels that have 5+ questions / day are fully answered proactively by AI, and people across departments go to the bot first for bug reports, incidents, and support tickets. Dozens of our beta customers have been consistently surprised by the quality of the answers, too.
Security is obviously super important for a product like this. We will never train on your data. In addition, your data is encrypted at rest, in an isolated database from other customers, and the system is SOC 2 compliant with regularly scheduled pen tests. We built it to Gather’s (https://gather.town/) SOC 2 Type II standards - that’s the original virtual office product we launched (and still maintain) out of YC, but we’ve since pivoted to Grapevine.
We put a lot of effort into making Grapevine easy to set up. You can try it now, for free, at https://getgrapevine.ai
21 comments
[ 5.6 ms ] story [ 40.7 ms ] threadWhat type of businesses are you targeting?
There is the ChatGPT product, operated by OpenAI, Inc, which you can access via their web site or their API. OpenAI does publish gpt-oss as an open-weights model. I suppose you could argue that gpt-oss is "a ChatGPT," though I'd normally think of it as "a large language model." Much like Claude, DeepSeek, Qwen and so on are other large language models.
do you have a very strong opinion about how companies should work?
"No"
Okay, does Dario Amodei? He thinks more than half the workforce should "just" be replaced. That's a strong opinion! Do you see what I am saying?
Do not assume that companies are willing to put ALL of their intellectual property into your hands. Even if you would not be some startup where any sysadmin could steal and sell my data any time without you even noticing it, you will get hacked just like everyone else that stores interesting data. The data you have access to is absolutely perfect for the global data blackmailing gangs. As soon as you are successful, you will have every black hat hacker and their dog knocking on your doors.
People can usually tell if an answer isn't helpful, but not always that it isn't accurate. Depending on the context, 85% accurate might not be good enough.
I was recently trying to tackle the same problem (@howie.systems). The hardest 2 problems we had to face were ACL and large files (and large volumes).
How did you solve the ACL part? I worked with a customer that had 200k pdf/images/dwg files on SharePoint and other 1M on samba. It took like a week to sync it all and keep tabs on the access rights of each employee.
How did you solve unpredictable large files: a pdf 2000pages, maybe some A0 in the mix. Or some 4GB power point presentations?
PS: great fan of gather. PPS: say hi to Clinton from me (amy.app) if he is still around. He was our mentor back in New Zealand at the flux accelerator (2016)
Forgive my cynicism, but $2 says they simply didn’t.
https://www.notion.com/product/enterprise-search