Teach the AI how you work

A skill is a reusable bundle of plain-language instructions about one concept, like a revenue definition, an analysis method, a team convention, or a chart preference. Each skill has a name and up to 50,000 characters of instructions. Basedash agents see the list of available skills and load a skill's full instructions only when a request is relevant, so you teach the AI something once instead of repeating it in every prompt.

Every agent sees the catalogue of skills, and when a request matches one, it loads that skill's full instructions before answering. You can check which skills shaped an answer in the thinking trace, where a step such as "Reading Revenue analysis skill" expands to show the instructions. Skills are picked up in chat, chart generation, dashboards, automations, and insights, and by Slack and MCP clients, which run the same agent.

Only organization admins can create, edit, or delete skills, which keeps shared playbooks consistent for everyone. Admins write them in Settings under AI context, ask the AI to draft or update one in chat, or manage them through the REST API. Members benefit from every skill without being able to change it, and can still shape their own answers with the personal AI context in their profile settings.

Global context is one block of text that every agent loads in full, so it suits a handful of durable facts, like what your company sells or what your team means by MRR. Skills are many named bundles, one concept each, and only the relevant ones load, so they suit longer playbooks. Group context sits in between: instructions for one team, such as finance's reporting conventions, applied to everyone in that group.

Yes, when it's worth keeping. Any member can ask the AI to remember a preference, and it saves it to that member's personal AI context. When an admin shares durable knowledge about the organization, the AI saves it at the narrowest scope that fits: a description on a data source, table, or column, global context, or a skill. Details that only matter to one conversation aren't saved, and everything the AI stores stays visible and editable, in settings or on the Data page.

What you say in the conversation wins. Explicit instructions in a chat take precedence over skills and saved context, so you can ask for bookings instead of GAAP revenue just this once without editing anything. Live data also beats notes: if saved context describes a column one way and the query results show otherwise, the AI goes with the results. To change the default for everyone, an admin updates the skill or the context.

Personal AI context shapes only your own answers, and Basedash agents are instructed never to copy personal information into organization-wide context. Shared context, such as global context and skills, can only be changed by admins, whether they edit it directly or ask the AI to. On Enterprise plans, AI edits to global context and skills appear in audit logs, and customer data is never used to train AI models.