Model-agnostic AI coworker that operates inside Slack and Microsoft Teams, connects approved tools, executes multi-step work and returns results in shared threads.
Product sources: OpenTag official product/pricing pages + Y Combinator S26 company page
OpenTag is a model-agnostic AI coworker designed to live in Slack and Microsoft Teams. A teammate tags it with work, it uses approved connected systems, selects a model for the task and reports finished work back into the originating thread.1 The current product also supports recurring jobs, a self-maintaining company wiki and cloud agents that run in isolated environments scoped to connected tools.1
OpenTag says its routing layer can reach more than 80 models rather than binding the workspace to one model vendor.1 Y Combinator's current S26 company profile independently corroborates the basic product identity: an active San Francisco company building an AI coworker for Slack and Teams that can answer questions, join calls, take on work and maintain company knowledge.2
The current trial is $0/month with a $100 credit grant, equivalent to 10,000 credits that do not expire, and does not require a card.3 Team costs $50/month and includes 5,000 shared monthly credits plus a one-time signing bonus matching the first month's credits; monthly credits reset, while the bonus does not expire.3
Enterprise pricing is custom and adds volume credit pricing, invoicing, SSO, SCIM, audit-log export, security-review support, a DPA, dedicated onboarding and custom retention.3 Credits are usage-based rather than seat-based, which makes access broad across a workspace but means effective cost depends on task mix and frequency.
OpenTag publishes unusually concrete controls for an early product. It says it reads only channels to which a human invites it, scopes integrations to the person who connected them, and waits for approval before sending, spending, submitting or purchasing on a user's behalf.4 Each customer workspace runs in an isolated environment; the security page states TLS 1.3 in transit, AES-256 at rest and separate credential storage with per-tenant keys.4
The current Terms, last updated August 18, 2026, make users responsible for delegated tasks and important actions and state that the service is provided as-is without a guarantee of uninterrupted, error-free or secure operation.5 The security page labels SOC 2 Type II as **in progress**, so AiToolMap does not describe OpenTag as SOC 2 Type II certified.4
AiToolMap's fixed 50-source panel did not surface eligible independent review or user-rating evidence for the current Open Curiosity product. That absence is neutral in the rating; it lowers confidence rather than counting as a negative score.
OpenTag can process substantial company and personal data. Its Privacy Policy, last updated August 28, 2026, says the service may handle connected Google Workspace data, Slack messages, files, contact details about third parties and, when users direct it to such content, sensitive or special-category information.6 It also documents that relevant task data may be sent to Microsoft Azure's model service, with contractual restrictions against model training, while connected Google user data is excluded from analytics and subject to Google's Limited Use requirements.6
The product's agentic risk is material because it can create, edit, send, share or delete content through connected services. OpenTag's mitigation is an approval model: the security page says sending, spending, submitting and purchasing wait for a person, permissions follow the user who connected the integration, and read-only connections cannot write.4 Customers should still apply least-privilege scopes, review retention and subprocessors, and treat approval policy design as part of deployment rather than relying on model accuracy alone.
OpenTag is best suited to teams already coordinating work in Slack or Teams and spread across systems such as GitHub, Stripe, Zendesk, Notion, PostHog, CRMs and internal databases.1 Its shared-coworker design is particularly relevant where several people need the same AI context and the same visible thread-level audit trail rather than separate personal assistants.
The product is less compelling for users who want a standalone desktop assistant, a self-hosted open-source stack or a predictable unlimited-use subscription. Buyers with heavy workloads should model credit consumption using their own recurring jobs and larger projects before assuming the $50 Team plan is sufficient.3
a clear team-chat interface; broad model routing; action-capable integrations; proactive recurring-job suggestions; a maintained knowledge layer; explicit approval gates for outbound/sensitive actions; granular connected-app permissions; a meaningful free trial; and detailed privacy/security documentation for a product this young.134
independent product evidence is currently sparse; SOC 2 Type II remains in progress; credit consumption can make monthly cost variable; the service depends on connected third-party systems and hosted model/infrastructure providers; and the Terms explicitly put responsibility for delegated tasks and review of important actions on the customer.5
Products matched by shared tasks, audience, workflow, product type and capabilities.