Grok Team Bots enter public beta with shared skills and private conversations

Grok's Team Bots are entering public beta on Teams and Enterprise plans, giving a group access to a shared bot built around a role or workflow. The September 28 announcement, published under SpaceXAI branding, describes bots with files, application connections and skills that colleagues can use together while retaining individual conversations.

The release's central idea is to provide a common starting context for work that otherwise requires each colleague to explain the same account, project or brand guidance again. The company illustrates that idea through its own deployments, rather than supplying independent evidence that every organization will achieve the same results.

Shared skills and individual context coexist

The publisher groups a Team Bot's inputs into context, plugins, credentials and memories. Context includes files, instructions and skills. Plugins connect applications such as Salesforce, Notion and GitHub, with connections configured by individuals or for the team. Credentials support third-party APIs that lack a plugin.

Although colleagues share the bot, the company says their individual conversations remain private, with separate context and memories for each person alongside shared skills. A bot can also join a Slack channel using its own handle, where participants can contribute context and see its replies. A shared channel and a private individual conversation should therefore be treated as different communication settings.

The announcement also describes learning from corrections and team decisions over time. It does not provide a complete specification for how administrators inspect, remove or restrict each kind of memory. A team considering sensitive information would need those controls documented before relying on the privacy description.

Account briefings and engineering coordination are the main examples

For sales accounts, the company says its bots review company news, Gong calls, Notion documents and Slack threads at night, then post morning briefings with role-specific drafts. Colleagues use the same bot to plan next steps and retain account decisions as personnel change.

The engineering example connects Slack with Notion, Linear, Hex, Datadog and Cursor. The bot follows decisions, triages reported bugs, creates tickets and launches cloud agents for defined fixes. Skills describe the team's shipping process and the evidence needed to consider a change complete.

The publisher reports using that setup to help a five-person team ship more than 100 pull requests per day while developing Team Bots. That is its account of its own process. A pull-request count does not establish the size, quality or business value of the changes, and it should not become a promised throughput target for another team.

The marketing example keeps a review step before launch: the bot checks drafts against brand material, produces previews for website changes and seeks final sign-off in Slack. That sequence is a more concrete description of oversight than assuming any connected agent can publish safely without an owner.

Data access depends on the permissions supplied

SpaceXAI's data example uses shared, read-only credentials for approved Databricks tables and connects other analytics tools. The team supplied a skills library developed over two years for working with more than 45,000 tables. This is substantial existing organizational knowledge, not evidence that a new bot can understand an undocumented warehouse on its own.

For an initial rollout, a narrowly defined role and a limited set of approved data sources would make the result easier to inspect. Teams should decide which connections permit actions and which are read-only, and name the person who approves changes. These are evaluation recommendations; Franklin has not tested the product's permissions or memory behavior.

The public beta makes the shared-bot model available to eligible plans and offers prebuilt examples for sales, product management, marketing and analytics. Its usefulness will depend on the accuracy of the supplied context and the limits around the tools it can use, not merely on how many colleagues can open the same bot.