Vybe vs Gumloop vs Claude Tag vs Grok Team Bots

Specialized coworkers with their own responsibilities and tools.

Reusable agents with skills and tool access.

Claude takes on tasks from a channel or DM.

Shared Team Bots built around a role, each with context, plugins, credentials, and memories; public beta on Teams and Enterprise plans.

Shared conversations and Slack threads with people and AI.

Shared agents with owner, editor, viewer, and use-only access.

Teammates can review and refine work in the same thread.

Everyone works with the same Bot and its expertise, while each person's conversations and memories stay private.

In-app chat, Slack, and email; other channels require setup.

Web chat, Slack, and API access to shared agents.

Slack channels and DMs with different access scopes.

Slack with a dedicated handle per Team Bot, alongside the Grok Bot app.

Delegate work to other AI coworkers.

Agents can hand tasks to other agents.

Not offered: one Claude works each thread; no delegation between specialized agents documented.

Bots pass context between each other, hand off tasks, and can pass ownership so you are not the router between tools.

Persistent memory, company knowledge, and reusable skills.

Reusable skills and playbooks that evolve with use.

Persistent channel context, memory, and standing instructions.

Team Bots keep separate context and memories per person while drawing on skills shared across the team.

Schedules, events, and follow-ups with run history.

Recurring or one-time scheduled runs plus event triggers from Gmail, Slack, Teams, and more.

Routines on a schedule, channel watch, or repository event.

Routines run on a schedule or, where supported, after an event; background routines run while the laptop is closed.

Hosted browser with persistent sessions; browser service required.

Interactive browser automation via the Website Scraper's Web Agent mode, an AI browsing node, and the Chrome extension.

Not offered: work runs in an ephemeral cloud sandbox through channel-scoped tool connections; no browser-control tool documented.

Every Bot runs on a persistent cloud computer with a browser, filesystem, and terminal, and can log in to services in that browser.

Supported meeting recording, transcripts, and follow-up; meeting service required.

No native recording or transcription; meeting work centers on Gumball pre-meeting briefs from a connected calendar.

Not offered: no meeting recording or transcription documented.

Not offered: connectors cover Gmail and Outlook calendars, but no meeting recording or transcription is documented.

Prebuilt connectors and custom integrations.

Connectors, shared credentials, and MCP support.

Admin-configured connections scoped to organizations, workspaces, or channels.

Built-in connectors, a preconfigured connector catalog, and custom MCP servers.

Build apps with AI, inspect code, and iterate in a live preview.

Not offered as a business-app builder: the platform ships agents, workflows, and hosted agent pages.

Interactive pages, charts, and code changes from the thread.

Bots steer Cursor Projects and launch Cloud Agents for code work; no interactive app-building surface documented.

Private, individual, and organization-wide app sharing.

Hosted agent pages on gumloopagents.com with sign-in and role-based access.

Hosted artifacts accessible to people in the channel.

Not offered: Bot share links are public previews on x.ai; no internal publishing surface documented.

Role-based access for owners, admins, editors, and members, with apps kept private, shared, or organization-wide.

Not offered: access control uses organization and team roles plus custom feature roles, not per-row data permissions.

Not offered: access is scoped per channel and by organization roles; no per-row data permissions documented.

Not offered: access follows connector permissions such as permission-aware Google Drive; no per-row data permissions documented.

SSH tunnels and custom connections for private resources.

Not offered: tools connect through cloud connectors, MCP, and webhooks; no private-network or SSH access documented.

Not offered: the ephemeral cloud sandbox reaches tools through configured connections; no private-network or SSH access documented.

Custom MCP servers must be reachable over the public internet; local servers require a tunneling service.

29 models across 11 providers.

35+ models advertised.

Choose from permitted Claude models, including Opus and Sonnet.

Runs Grok models; no cross-provider model catalog documented for Bots.

Auto routes tasks to Cheap, Balanced, or Smart tiers.

Auto picks the model for each agent message, with mid-run escalation and fallbacks.

Choose a model for a thread, channel, or DM; automatic task routing not verified.

Not offered: no automatic task-to-model routing documented.

Choose tier models, map task categories, or pin a model.

Admins restrict models, pin per-agent presets, and set fallbacks; task-category mapping not offered.

Admins set workspace and channel defaults; individual threads can switch models.

Not offered: no task-category-to-model routing rules documented.

Separate conversation and coding routing policies.

Not offered: AI model governance restricts and pins models platform-wide; no separate coding policy documented.

Not offered: admins choose permitted Claude models per scope; no separate coding policy documented.

Not offered: Cursor team rules and privacy settings apply to Bots; no separate coding model policy documented.

Bring your own provider keys as part of an enterprise setup.

BYOK on Pro and above; organization-wide keys on Enterprise.

Not offered: Claude uses admin-provided tool accounts with credentials kept out of the sandbox; no customer model keys documented.

Credentials connect third-party tool APIs; no customer AI provider keys documented.

OpenRouter and Vercel AI Gateway.

Provider proxy URL, custom headers, and model name mapping per provider on Enterprise.

Not offered: no bring-your-own model gateway documented.

Not offered: no bring-your-own AI gateway documented.

SAML single sign-on for enterprise.

SAML SSO and SCIM for enterprise identity.

Claude Team and Enterprise organizations support SAML SSO and JIT provisioning via WorkOS.

Members sign in with their Cursor account, so existing Cursor SSO and team membership apply.

Access, agent activity, and tool usage for enterprise review.

Agent activity, tool-call tracking, and analytics.

Documented activity trails; review your required audit scope.

Advanced audit controls ship on Enterprise plans; Grok Bot-specific audit coverage not documented.

Action-specific permissions and contact controls on every channel; explicit approval gates not documented.

App guardrails, spending approvals, and scoped policies.

Every action passes a permission checker; admins pre-approve action types per scope with auto-mode allow rules.

Require Approval rules and Auto Review evaluate tool calls and computer actions; passwords and 2FA use human takeover.

Usage visibility and per-agent credit budgets.

Spend caps by agent, team, or organization.

Monthly organization usage limit; DMs use personal accounts separately.

Pooled billing with weekly usage allowances on premium seats; per-Bot budget controls not documented.

Free plan with $10 in credits every month; paid plans from $20/month. $1 = 100 credits; top-ups available.

Pro starts at $37/month; credits and an orchestration fee apply.

Channel work draws from the organization’s usage balance.

Individual access via SuperGrok Heavy or Cursor Ultra; self-serve teams use premium seats with usage allowances; Enterprise is rolling out.

Unlimited builders and viewers in your organization.

Unlimited seats listed on Pro and Enterprise.

Channel usage is not charged against individual seats; account eligibility still applies.

Team seat management and consolidated billing on paid plans.