2.7 Base in Depth: Views, Fields, and Sharing Permissions
Lesson 1.8.2 covered what Base is and why it exists. But that only used the most basic layer: put data in, let the agent query it.
Using Base seriously raises three further questions: which view should I use? How should fields be designed so the data stays usable? Who can change which table? This lesson takes them in turn.
Choosing the Right View
The same data, seen from a different angle, answers a different question. Pick the wrong view and you will stare at a wall of numbers without finding the point.
| View | Best at answering | Typical use |
|---|---|---|
| Table | Reviewing row by row, bulk editing, checking detail | Reconciliation, data cleaning, filling in missing fields |
| Kanban | Grouping by status, seeing where work is stuck | Invoice processing, résumé screening, ticket flow |
| Gallery | Browsing by image | Product galleries, design assets, sample photos |
| Assets view | Reviewing linked file assets | Product or material records with attachments |
Start with 【Table】 to check details. Use 【Kanban】 to follow status, 【Gallery】 to browse images, and 【Assets view】 to review linked file assets.
A view is a different rendering of the same data, not a copy of it. Change a value in one view and every other view updates — which is precisely where it beats keeping a separate spreadsheet per scenario.
For creating views, filtering, sorting, and grouping, see Views and Filters.
Field Types and Validation Rules
Field design decides how long the data stays usable. A few practical principles:
| Principle | Why |
|---|---|
| One record holds one entity | Mixing customers and orders in one table forces an extra filter on every later count |
| Use a date field for time, not text | Date fields sort and filter as dates, avoiding mixed text formats |
| Split amount into "value" and "currency" | Merged together, it cannot be aggregated or compared across currencies |
| Use a select field for status, not free text | Free text produces "Done / done / OK" as three different values |
| Define the meaning of any field the agent writes to | If the field is vague, it fills in its own interpretation |
That last point matters most. When the agent extracts data, the field name and its description are the specification it writes against. A well-named field with no description can still end up with the delivery date in the order-date column.
For creating fields and linking Library resources, see Tables and Fields. For adding and editing data, see Record Management.
Permissions and Sharing: Who Can Change Which Table
Base permissions come in two layers, and they are easy to conflate:
| Layer | What it controls | Who can change it |
|---|---|---|
| Entry authorization | Whether you can open Base at all | You (completed in 【Explore】, see 1.8.2) |
| Data permissions | Which tables and records you can see or modify | The resource owner or an administrator |
That is why "Base opens, but one table is missing" happens. It is not a fault; the two layers simply govern different things.
The three system roles:
| Role | What they can do | Typically assigned to |
|---|---|---|
| Owner | Full control, including permission settings | The data owner |
| Admin | Read, edit, and manage the data | Core team members |
| Reader | Read-only access | Colleagues who need to see but not change |
Roles and base-level permissions are configured in AOS Studio. In AOS Work, view and edit within your assigned permissions.
A practical habit when opening access: start everyone as Reader. Promote to Admin only where collaborative editing is genuinely needed, and keep the number of Owners small.
For adding members and assigning roles, see Create and Manage a Base. For querying, updating, and reading records again in a conversation, see the Base exercise.
How Base Divides Work With Library and Memory
The boundary between the three data carriers is worth restating:
| Carrier | What it holds | How the agent uses it | The test for putting something there |
|---|---|---|---|
| Library | Source files | Reads, references, rewrites | It is "a document" |
| Base | Structured records | Queries, filters, writes, updates | It is "records that can be compared one by one" |
| Memory | Your preferences | Carried across tasks | It is "a long-term fact about you" |
A common misuse: drop a 200-row spreadsheet into Library and ask the agent to "find everything with stock below 200". It first needs to read the file content. In Base, the same question can be queried repeatedly against the fields. If you will ask the same kind of question repeatedly, move it into Base.
Troubleshooting
| Symptom | Cause | What to do |
|---|---|---|
| The agent says it cannot find the table | The table is outside what it can reach | Confirm entry authorization, and ask the owner to open that table |
| Written records land in the wrong fields | Field names or descriptions are unclear | Add field descriptions, or state the field mapping explicitly in the task |
| Data conflicts after concurrent editing | No division of labour | Split by record owner, or mark rows as In Progress with a status field |
Summary
Base's depth is not in how many features it has, but in whether the data is designed to last. Views decide how you look, fields decide whether it can be calculated, and permissions decide who can change it.
That completes the data layer, and with it both the method and asset lines. The next lesson is the appendix: the whole system on one page.
2.6 Library in Depth: Team Space, Bulk Work, and Knowledge Sources
Lesson 1.8.1 got your files in. This one covers what happens once you have thousands of files and a whole team.
2.8 Appendix: How the AI Work System Runs
Explains what happens underneath a task: the role and limits of the model, context, prompts, the agent, tools, Skills, MCP, knowledge bases, and workflows.