[BUG]: Context tab shows Context Limit / Usage as em-dash for custom openai-compatible models even with configured limit.context #37456

Description

The Context tab (the token/context breakdown panel opened from the circular context-usage indicator) shows — for the "Context Limit" and "Usage" stats when using a custom @ai-sdk/openai-compatible provider — even when an explicit limit.context is configured for that model in opencode.json.

Because the limit renders as —, the usage percentage is also —, so the whole context-tracking readout is unusable for these models, and there is no visible indication anywhere of the model's context window.

Root cause (from source)

packages/app/src/components/session/session-context-metrics.ts:

type Model = {
  name?: string
  limit: {            // declared non-optional…
    context: number
  }
}

const build = (messages, providers) => {
  const message = lastAssistantWithTokens(messages)
  if (!message) return undefined
  const provider = providers.find((item) => item.id === message.providerID)
  const model = provider?.models[message.modelID]
  const limit = model?.limit.context          // …but accessed as if it can be missing
  const total = tokenTotal(message)
  return {
    ...
    limit,
    usage: limit ? Math.round((total / limit) * 100) : null,
  }
}

limit resolves to undefined whenever:

1. provider.models[message.modelID] returns no entry (the model the SDK reports for the assistant message is keyed differently than the config entry, e.g. gateway/router providers), or
2. the resolved model object has no limit (custom/local models whose window can't be auto-fetched and no override was merged into the resolved model).

Note the type declares limit: { context: number } as required, but the code reads model?.limit.context. If a resolved model exists but has no limit, model.limit is undefined and .context is an unsafe access. Either way the tab falls back to the em-dash.

packages/app/src/components/session/session-context-tab.tsx:

{ label: "context.stats.limit", value: () => formatter().number(ctx()?.limit) },
{ label: "context.stats.usage", value: () => formatter().percent(ctx()?.usage) },

session-context-format.ts renders undefined/null as "—", so a missing limit surfaces as — with no fallback.

Expected behavior

The Context tab should display the model's context limit whenever it is knowable:

1. If the user configured limit.context in opencode.json, always display and use it (config override should win over auto-detected metadata).
2. If not configured, fall back to models.dev / the provider's /v1/models metadata.
3. If still unknown, show a clear hint (e.g. "not set — configure limit.context") instead of a bare —, so it's discoverable.
4. model.limit should be treated as optional in the type (limit?: { context?: number }) since custom/local providers legitimately lack it, and the access should be null-safe.

Actual behavior

"Context Limit" and "Usage" render as — in the Context tab; no context window is shown anywhere for the model.

Steps to reproduce

1. Configure a custom OpenAI-compatible provider with an explicit limit:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "omniroute": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "OmniRoute",
      "options": { "baseURL": "http://localhost:20128/v1", "apiKey": "sk-..." },
      "models": {
        "hyperspaceaihai/anthropic--claude-opus-latest": {
          "name": "opus",
          "limit": { "context": 1000000, "output": 128000 }
        }
      }
    }
  }
}

2. Start a session on omniroute/hyperspaceaihai/anthropic--claude-opus-latest and send a few messages so token usage accrues.
3. Open the Context tab from the circular context-usage indicator.
4. Observe "Context Limit: —" and "Usage: —", despite limit.context: 1000000 being configured.

OpenCode version

1.18.3

Operating System

Windows 11 (10.0.26200)

Related issues

• OpenCode Desktop shows Context Limit = 0 for LM Studio providers. #35955 — Context Limit = 0 for LM Studio (auto-fetch of /api/v0/models metadata; this report is broader: explicit config override also not shown)
• fix: context limit display uses stale model after model switch #35072 — context limit display uses stale model after model switch
• [FEATURE]: Feature Request / Bug: Cannot set context window limits for local models via GUI/TUI (Defaults to 0) #31433 — request to set limits via GUI/TUI (this issue is about displaying a configured limit)
• [BUG]: Custom openai-compatible provider model limit overrides ignored for context percentage display #29555 (closed) — TUI status-bar percentage ignored configured limit.context (same class, different surface: status bar vs. Context tab)
• [BUG]: Context percentage indicator doesn't reflect usable context proportion #28057 — context percentage indicator doesn't reflect usable context proportion

Filed by an AI assistant at the user's request after inspecting the config, the session-context-metrics.ts / session-context-tab.tsx / session-context-format.ts source at HEAD, and the existing issue tracker.