Work with Channels

A Channel is a shared place for related business context. It can bring together Data Sets, files, source information, and Sessions without changing which source systems remain authoritative.

Structured channel data follows one hierarchy:

Channel
└── Data Set — owns one schema and its records
    ├── View — renderer + filters + display configuration
    ├── View
    └── View

A Data Set is the single home for one kind of record and its fields. Views do not create another schema or copy the records: they present the same Data Set as a Table, Board, Calendar, Timeline, Dashboard, or Pivot table. This means a field change applies consistently to every View over that Data Set.

Open a Channel

Choose Channels from the main navigation, then open a Channel you can access.

The Channel first shows its Data Sets. Open one to work with its records and choose one of its saved Views. Not every Channel needs structured data, and a Data Set only needs the Views that are useful to its team.

Choose the right view

Saved views help a team return to the same useful perspective instead of rebuilding filters each time.

Filters in the toolbar are temporary until you save them. Choose Save to replace the current View after a version check, or Save as to create a new named View and leave the current one unchanged. A saved View includes its renderer, filters, and display configuration; it continues to use the Data Set's one shared schema and records.

Pivot tables are designed for rolling financial statements and other cross-tab reports. A date field can create live month, quarter, or year columns, while measures or a line-item field form the rows. New source periods appear without adding a new physical field to the source schema. Missing values remain visibly empty; they are not silently converted to zero.

For reviewed reporting, a Pivot table can exclude the current incomplete date bucket without fixing an end date. The view then rolls forward automatically when that period becomes complete.

Dashboard notes can either preserve a reviewed snapshot or be generated from named metric and chart widgets. Generated notes are recalculated from the same live dashboard results whenever the view loads, so period labels, values, and comparisons roll forward with the source data. Their trace identifies the exact widgets used; they are not a second hidden query or a saved model narrative. Date charts can also exclude their current incomplete day, week, month, or year without fixing an end date. Notes generated from those charts inherit the same rolling boundary, so a reviewed dashboard does not summarise a partial period.

Work with records

Records represent the structured items available in a Channel. Depending on your access and the source of a field, you may be able to create, update, or link records.

Some fields are owned by a connected source and cannot be edited directly in the Twin workspace. This protects the original system from accidental changes and makes the source of a value clear.

Where available, provenance details show where a record or value came from.

Understand people and agent activity

Channel posts, linked Sessions, and artifact changes can originate in the web app or through the CLI, API, MCP, Codex, Claude Code, and other approved clients. The activity feed places provenance on a quiet sub-row beneath the item instead of mixing it into the content.

The authenticated person’s avatar and initiating client badge overlap as a pair. The Twin mark can join them when governed Twin data was actually used. Hover or focus the group to read a standard explanation such as “Archie Norman posted this using Claude Code via CLI.” The same compact explanation appears in the Channel sidebar preview. Provenance is an audit and display signal, not an authority claim.

Use supporting context

A Channel may also include:

• Sources, showing connected evidence relevant to the Channel;
• Memory, holding useful shared context for work in that Channel; and
• Files, containing documents available to permitted users and Sessions.

Channel memory helps ongoing work, but it is not the same as published Twin knowledge. Published knowledge has its own review, versioning, and evidence.

Start work from a Channel

Start a Session with the Channel or one of its saved views attached when the agent should work from that context.

This is more precise than asking the agent to search across unrelated parts of the workspace, and it keeps the resulting work linked to the place where the team expects to find it.