Every team chat has the same problem. Someone asks a question, someone else half answers it, and the real work happens in six other tabs. OpenTag's bet is that the fix isn't another dashboard. It's an AI agent you tag right in the thread, like a colleague, and it hands the result back where everyone can see it.
That pitch landed OpenTag a spot in Y Combinator's Summer 2026 batch. It also puts the company in a crowded, fast-moving corner of enterprise AI, up against Anthropic's Claude Tag, which works the same way but only with Claude models. So the question isn't whether the idea is good. It is. The question is whether a three-person startup has shown enough yet to earn a place in your Slack workspace.
The essentials
| Detail | What we know |
|---|---|
| Company | OpenTag (tryopentag.com) |
| Founded | 2026 |
| Headquarters | San Francisco, California |
| Founders | Tony Kam (CEO), Shelden Shi (CTO), Wilson Nguyen (Chief Engineer) |
| Backing | Y Combinator, Summer 2026 (S26) |
| Product | AI coworker for Slack and Microsoft Teams |
| Main features | Model-agnostic agent, auto-generated company wiki, workflow automation, approval gates |
| Integrations | Slack, Microsoft Teams, HubSpot, Gmail, Stripe and others |
| Billing | Monthly credits consumed per task |
| Team size | 3 (the co-founders), as of mid-2026 |
| Previous venture | Lilac Labs (YC S24) |
| linkedin.com/company/tryopentag | |
| Icon Polls rating | 3.0 out of 5 |
Second time through Y Combinator
The founders aren't newcomers. All three studied at UC Berkeley and worked together at Lilac Labs, their earlier YC company from the S24 batch, before pivoting into what became OpenTag. Kam came from Tesla and Intel. Shi worked at Flatiron Health and Roche. Nguyen came from Hewlett Packard Enterprise.
That matters more than it might seem. A team that has already been through an accelerator, built something, and decided to change course usually has fewer illusions about how hard the next part is. It doesn't guarantee a good product. It does make the company easier to take seriously.
Tag it, give it a job, read the result
The agent, first launched under the working name Gini, is meant to do more than answer questions. According to OpenTag's own materials and its YC profile, it can take on multi-step tasks, connect to tools like HubSpot, Gmail and Stripe, and even join calls. Each task runs on a sandboxed cloud machine, and sensitive actions sit behind approval gates that need a human sign-off.
The headline technical claim is that OpenTag is model-agnostic. Rather than tying itself to one AI provider, it says it routes each task to whichever of roughly 80 models suits the job, and it advertises savings of up to 70% against single-model rivals. Those are the company's figures, not independently verified ones.
The logic is sound. When a better model ships from any lab, users should benefit without switching tools, and the company isn't hostage to one vendor's pricing. But there are two catches. Public documentation on how routing decisions get made is scarce, and that is exactly what an enterprise buyer will ask about first. And an assistant that quietly swaps models between tasks can produce answers that vary in quality and tone in ways users notice.
The wiki that writes itself
This is the most interesting idea in the product. As OpenTag works, it observes how the company operates and builds a knowledge base: who owns what, how decisions get made, which workflows keep repeating. The aim is documentation that updates itself instead of going stale the week after someone writes it.
Early adopters on Product Hunt singled this feature out for praise. That feedback was positive but thin, with few detailed case studies or benchmarks to point to. For a company this young, that's normal. It still means the auto-wiki is a promise more than a proven result.
Pricing you have to ask about
OpenTag sells monthly credits, and each task eats some of them. That's a common model for AI tools in 2026. The problem is that the cost per task wasn't clearly published on the website when this review was written, and there was no free trial or sandbox to try first. The main route in is booking a demo, which points to a high-touch sales process rather than self-serve sign-up.
The Product Hunt launch offered 50% off for two months. Nice, but discounts on an unclear price are hard to judge. Credit billing only works when teams can predict what a typical week will cost them.
Not every OpenTag is this OpenTag
Search for the name and you'll run into a lot of noise. The commercial product appears to be closed source, with no public repository, SDK or developer docs on GitHub. What you will find there belongs to others: an open-source organisation called OpenTagOS working on PHP tooling, Terraform providers and testing frameworks, plus several community projects, including forks by CopilotKit, that borrowed the name for open-source agent-mention workflows in Slack and GitHub.
LinkedIn has the same issue. Alongside the startup's own young page (search for tryopentag), there's a Malta technology services firm founded in 2014 and a larger iGaming platform provider in Sofia, Bulgaria. Glassdoor reviews filed under OpenTag belong to that Sofia company, not this one. For a startup trying to build trust, that kind of confusion is a real cost, and the lack of any public code will bother technical evaluators.
Jobs, for now, are a waiting game
There were no public openings tied to the AI coworker product at the time of writing. YC and Dealroom both list the team at three, and Dealroom puts the estimated enterprise value at $500,000 to $750,000, a figure that should be read as an estimate, not an official valuation. About 10 teams were in early access at the YC launch.
Given the founders' engineering backgrounds, early hires will likely lean toward engineering and product. Reaching out through LinkedIn or the contact details on the YC profile is the sensible route. Just don't confuse listings from the Sofia iGaming company with this one.
Our verdict: 3 out of 5, for now
- Worth liking: working inside Slack and Teams removes friction, the model-agnostic approach makes strategic sense, the auto-wiki tackles a genuine headache, and the founders have real pedigree.
- Worth waiting on: clear pricing, security and routing documentation, a trial option, and evidence from more than a handful of teams.
OpenTag feels like a product still being shaped. Early adopters who are comfortable with ambiguity may get value now. Everyone else should keep an eye on it. If you've got opinions on which online tools actually earn their place in your day, add your vote to our poll on the world's most useful websites.
Is OpenTag the same as Claude Tag?
No. Claude Tag is Anthropic's Slack teammate and runs only on Claude models. OpenTag is a separate company offering a similar tag-in-thread experience that routes tasks across many models.
Is OpenTag safe for sensitive data?
The company says each task runs in a sandboxed cloud machine that is torn down afterwards, with human approval gates on sensitive actions. Detailed security documentation isn't public, so teams handling sensitive data should request a full security review first.
Is OpenTag open source?
The commercial product at tryopentag.com isn't. The open-source projects using the OpenTag name on GitHub are unrelated to the YC-backed company.
