Context is shared memory for an organisation and its agents. It helps the agent understand the business, use the right data, keep calculations consistent, and avoid recommending work that has already been tried.
Open it from Agent → Context.
How Context is organised
The Context page uses tabs. You do not need to know the internal model names; start from what you are trying to do.
| Intent | Context tab | Use it for |
|---|---|---|
| Understand the organisation | Business Model | Customers, channels, value proposition, activities, costs, and revenue model |
| Connect business concepts to data | Data Map | Entities, events, relationships, datasets, and column mappings |
| Keep calculations consistent | Feature Catalog | Metric and feature definitions, calculation logic, owners, and linked assets |
| Track worthwhile opportunities | Use Cases | Ideas for dashboards, analyses, reports, models, or decisions |
| Set preferences and rules | Conventions | Naming, style, modelling, reporting, and analytics preferences |
| Teach repeatable methods | Skills | Reusable playbooks for specialised or recurring agent work |
Overview
The Overview tab explains the purpose of each context area and gives a simple rule of thumb:
- Business facts go in the Business Model
- Data meaning goes in the Data Map
- Calculation rules go in the Feature Catalog
- Opportunities go in Use Cases
- Preferences go in Conventions
- Repeatable procedures go in Skills
Posted versus trusted knowledge
Some context can be proposed by humans or agents, then reviewed before it becomes trusted organisational knowledge. Use Agent → Evaluations to review proposed or inconsistent context when review workflows are enabled for your role.
This matters because agents will use trusted context when recommending actions, writing reports, building dashboards, or creating jobs.
Examples
Good context entries are specific enough to change future work:
Our primary customer segment is mid-market professional services firms with 50-500 employees.
Monthly recurring revenue excludes implementation fees and one-off services revenue.
When reporting pipeline conversion, use opportunity_created_at for cohorting and closed_won_at for conversion timing.
For board reporting, use concise commentary with no unexplained statistical jargon.
Best practices
- Add durable facts, not passing chat thoughts.
- Prefer one clear entry over a long, mixed note.
- Link context to datasets, dashboards, reports, or pipelines when the relationship matters.
- Review agent-proposed changes before treating them as official.
- Update context when the business changes, not only when a dashboard breaks.