Context usage

In DataMind Oda, every chat works inside the model's reading space, and the chat's work fills that space as it goes. The context ring in the message box shows how full it is; this page covers the ring and the panel it opens. Chat with Oda covers the chat itself.

The context ring

The ring sits in the message box's control row, just before the model picker. The message box covers the row and its other controls.

• The ring shows the share of the model's context window that is in use, as a ring and a percentage; point at it to read {percent} of context used.
• It stays hidden until both the chat's usage and the model's capacity are known; a fresh chat shows no ring.
• The percentage follows the chat's latest figure: it moves as the chat fills, and it drops at once when an older part of the chat is compacted.
• When the control row runs out of space, the ring gives up its percentage first and keeps only the ring itself.

The breakdown panel

• Click the ring to open the breakdown panel. Click the ring again, press Escape, or click outside the panel to close it; a click inside the panel does not close it.
• The panel reads the percentage of the context in use, and the chat's figure against the model's limit, shown in a compact form such as ~32K / 128K.
• Below the figures, three rows follow the bar's order: System prompt, Tool definitions and Messages, each with its share of the space.
• Only the percentage is exact. The figures and the split below it are an estimate, which the ~ marks.
• A part that takes no space of its own draws no segment in the bar. If the part-by-part split is not available at all, the panel shows the figures without the three rows.
• If a switch to another model leaves the capacity unknown, the panel closes; it stays closed after the capacity returns, until you open it again from the ring.

When the context fills

When a long chat no longer fits the model's reading space, Oda shortens the older part of it, on its own or when you run /compact, and the ring follows the new figure at once. The marker the compaction leaves in the conversation, and its summary, are covered in System prompt, context and compaction.

Questions, answered

What do the numbers in the panel mean?

The percentage is the share of the context in use, like 25% of context used; beside it, the chat's figure stands against the model's limit, like ~32K / 128K. The three rows below split that figure between the system prompt, the tool definitions and the messages.

Why do some figures start with a ~?

The ~ marks an estimate. Only the percentage is exact; the figures and the split beside it are estimated from it, so they carry the ~.

What do the three rows stand for?

They split the context by what fills it: the instructions Oda sends (System prompt), the definitions of the tools it can use (Tool definitions), and the chat's messages (Messages).

Why did the panel close by itself?

The panel closes when the model's capacity is no longer known, for example after a model switch that leaves it unknown. It stays closed even after the capacity returns, until you open it again from the ring.

What happens to the ring when a long chat is compacted?

Oda shortens the older part of the chat, and the ring follows the new figure at once, so the percentage falls. The marker the compaction leaves, with its summary, is covered in System prompt, context and compaction.

Why does the ring sometimes show without a percentage?

When the control row runs out of space, the ring gives up its percentage first and keeps only the ring itself. The Oda window, sidebar and panels covers how the window folds and gives way when it is narrow.

Why does the panel sometimes show no rows?

The three rows appear when the part-by-part split is available. Until then, the panel shows the percentage and the figures without them.

What does the number after the slash mean?

It is the model's context window: the most text the model can read at once. The figure before the slash is how full that space is.