OpenTag Review (2026): Pricing, Features & Honest Verdict

OpenTag is an AI coworker that lives in Slack and Microsoft Teams, joins the channels you invite it to, and finishes multi-step work across your connected tools instead of describing it. Best for: small ops-heavy teams already running on Slack. Price: from $50/mo, with a $100 free trial and no card. Rating: 7.5/10.

What is OpenTag?

OpenTag is a shared AI agent built by Open Curiosity, Inc., a San Francisco company backed by Y Combinator. Three cofounders run it: Tony Kam on product, Shelden Shi on engineering, and Wilson Nguyen on infrastructure and security. All three came out of Lilac Labs (YC S24) and Berkeley, with earlier stints at Tesla, Intel, Flatiron Health and HPE. That matters here because the product is a sandboxed machine with credentials to your company tools, and you want to know who built the permissions layer.

The pitch is narrower than most agent products, and better for it. You @mention OpenTag in a Slack channel the way you would mention a colleague. It reads the channel, plans the job, runs it on its own cloud machine across the tools your team connected, and posts the finished artifact back to the thread. Not a checklist. The report, the file, the updated CRM record. We found the docs unusually blunt about the boundary: “It is not a search box over your Slack. It reads to work, not to index.”

What are OpenTag’s key features?

Jobs that run on their own machine

Every job gets a sandboxed cloud machine scoped to the integrations you connected, and that machine is destroyed when the run finishes. This is the difference between OpenTag and a chat window. It can spend twenty minutes on a task, touch six tools in one run, and keep working after you close Slack. The docs give a worked example that is refreshingly concrete: OpenTag paused a retargeting campaign taking 38% of ad spend at 3x blended CAC, moved the budget, and reported roughly $1,240 a month recovered.

Approval gates enforced by the runtime

Before anything irreversible reaches another person, OpenTag stops and asks. Sending an email, submitting a form, spending money. The security page is specific that this waiting is enforced by the runtime rather than by the model deciding to be careful, which is the right architecture and a claim most agent vendors avoid making. Integrations are also scoped per person, so OpenTag uses your Stripe access only when you say so, and a connection you grant read-only cannot write at all. GitHub and Granola are read-only with no configuration needed.

Routines it offers to take over

Nobody sits down and configures workflows. When OpenTag notices the same request three Mondays running, it offers to own it, and you answer yes or no in the thread. That is a smarter onboarding path than a builder canvas, because the automation gets proposed from real usage instead of guessed at on day one.

A company wiki that writes itself

OpenTag builds pages out of how the company actually operates, sourced back to the threads they came from. Ask why the enterprise tier got dropped and you get a sourced answer with the decision and the person who made it, because OpenTag was in the room. It keeps those pages current as decisions change. Good idea, and the one feature here we would stress-test hardest before trusting it, because a stale wiki that looks fresh is worse than no wiki.

How much does OpenTag cost?

OpenTag starts at $50 a month for 5,000 credits, and the trial is a $100 grant of 10,000 credits with no credit card and no sales call. Every plan gets every skill, every integration, and unlimited channels and people. There is one OpenTag per workspace and everyone shares it, so adding your whole team does not change the price.

Here is the part the pricing page does not tell you. The Team card shows a single $50 price, but the docs publish a five-rung ladder: 5,000 credits for $50, 10,000 for $100, 20,000 for $200, 50,000 for $500, and 100,000 for $1,000. Do the division and every rung lands on exactly $0.01 per credit. There is no volume discount anywhere on the published ladder, even though the Enterprise card advertises “volume credit pricing with discounts.” Committing to twenty times more credits buys you nothing per unit. The only real discount is annual billing at 15% off, and that one reconciles to the cent on all five packs: $90, $180, $360, $900 and $1,800 saved per year.

The bigger issue is what $50 actually buys. OpenTag’s own docs price a weekly report at about 375 credits per run, which is roughly 1,500 credits a month. So the entry plan covers about three weekly routines and nothing else. Then the docs walk through an example team running four weekly routines, a daily triage sweep, 200 ad-hoc questions and a few larger analyses, and total it at about 25,000 credits a month. That is $250, five times the advertised entry price. The $50 headline is real, but it is a starter rung, not the plan a working team lands on.

Against per-seat competitors the maths can still favour OpenTag. Zapier charges $19.99 a month for Professional and $69 for Team, and Slack’s own AI is bundled into paid Slack plans you are already buying seats for. If fifty people ask OpenTag questions, you pay for the work, not the fifty seats. If three people use it heavily, per-seat pricing wins.

Who is OpenTag best for?

Use OpenTag if your team already runs on Slack or Teams, you have recurring operational work that crosses three or four tools, and the bottleneck is that nobody has time to do it rather than that nobody knows how. Ops, finance, support and growth teams at companies between ten and a hundred people are the obvious fit. The credit model rewards you for having a lot of occasional users, which is exactly the shape of a small company.

Skip OpenTag if your stack is not on the connector list. There are fourteen built-in integrations today: Airtable, Attio, Bland AI, GitHub, Gmail, Google Calendar, Google Docs, Google Drive, Google Sheets, Granola, HubSpot, Linear, Notion and PostHog. Note what is missing. The homepage shows Stripe and Zendesk logos, and the security page literally uses Stripe as its example of scoped access, but neither is a built-in connector. You can reach them through a custom MCP server or a generic API key, and OpenTag is honest that this is its answer to most “do you integrate with X” questions. It is still more setup than the marketing implies.

Also skip it if you need a compliance certificate today. SOC 2 Type II is listed as in progress, not complete.

What are the best OpenTag alternatives?

Slackbot and Slack AI. Slack’s own agent is native, free of extra setup, and bundled into paid Slack plans. It is a personal assistant grounded in what you can already see, good at finding, summarising and planning. OpenTag publishes its own comparison page and is straight about the split: Slackbot is personal, OpenTag is shared execution across tools. Many teams will run both.

Zapier. Free for 100 tasks a month, $19.99 for Professional, $69 for Team. Zapier has thousands of connectors against OpenTag’s fourteen, and it is the right answer when you know exactly what should happen and just need it wired. It is the wrong answer when the job needs judgement, because a Zap does not read a messy thread and decide what the ask really was.

Dust or Glean. Both sell company-context AI to larger organisations. Neither publishes a self-serve price, which tells you where they aim. If you have a procurement process and a security questionnaire, they belong on the list. If you want to be running something by Thursday, they do not.

Final verdict: is OpenTag worth it?

OpenTag is the most convincing version of the Slack-agent idea we have tested this month, mostly because the team resisted the urge to make it do everything. The scoping story holds up: invited channels only, per-person credential scoping, approval gates enforced by the runtime rather than the model, no training on your data as a contractual term rather than a toggle. The documentation is better than the marketing site, which is rare and a good sign about where the effort went.

The pricing is where we would push back. A perfectly flat $0.01 per credit across a 20x range, sitting next to an Enterprise card promising volume discounts, reads as a ladder that has not been thought through yet. The $50 entry price is technically true and practically misleading, since the vendor’s own example workload costs $250. And the connector gap between the logos on the homepage and the fourteen tools in the docs should be closed. Buy it if you are on Slack, have real recurring work, and are willing to budget $150 to $250 a month rather than $50. Spend the $100 trial on one properly annoying weekly job before you commit, because that single measurement tells you more than any of this.