TUI context usage always shows 0% — model.limit.context is undefined

[wanghaozi]

Description

The context usage indicator in the TUI (the progress circle button showing "used" percentage in the right panel) always shows 0%, regardless of how much conversation has occurred.

Root Cause Analysis

The issue is in the build\ function (which computes session context) in the bundled frontend code. The chain is:

1. build\ computes usage from the model's context limit:

   const limit = model?.limit.context;   // line 130424
   usage: limit ? Math.round(total2 / limit * 100) : null  // line 130435

2. When model.limit is undefined (server doesn't return it), limit becomes undefined, usage is set to null.

3. SessionContextUsage component renders:

   percentage: context()?.usage ?? 0   // null → 0

4. ProgressCircle shows 0%:

   const clampedPercentage = Math.max(0, Math.min(100, 0 || 0));  // = 0

Two Possible Causes

1. Server API doesn't return limit.context — The model.limit field at line 75945 is passed straight through from the server's model list API response (sdk.model.list()). If the server doesn't include limit: { context: <number> } in the model data, limit is undefined.

2. Server doesn't send token data in session.step.ended — At line 78140, event.data.tokens populates the assistant message. If this is empty, tokenTotal returns 0, lastAssistantWithTokens skips the message, and build\ returns undefined → 0%.

Additional crash risk

The same model.limit is accessed without null safety in ModelTooltip component (line 136394):

props.model.limit.context.toLocaleString()  // crashes if model.limit is undefined

Suggested Fix

In build\, add a fallback for model.limit:

const limit = model?.limit?.context ?? 0;

And in ModelTooltip:

props.model.limit?.context?.toLocaleString() ?? "unknown"

But the real fix should be on the server side to ensure model.limit.context is always populated with the correct context window size for each model.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [BUG]: Context tab shows Context Limit / Usage as em-dash for custom openai-compatible models even with configured limit.context #37456: Same root cause — model.limit.context is undefined in session-context-metrics.ts, causing context usage to show as — / 0% (closed, but covers the same unsafe access and null-safety fix)
• Token counter broken in Desktop App with Synthetic provider (cost stuck at $0, usage at 0%, tokens frozen) #39092: Context usage stuck at 0% with Synthetic provider — overlapping symptom (usage shows 0%) attributed to a provider not returning token metadata, which intersects with your second cause (session.step.ended tokens being empty)