Summary

Fixes the Ollama bridge metadata for current GPT-5.5 / GPT-5.4 models so /api/show reports a 400k effective context window instead of smaller stale fallback values.

This updates the bridge context override table for:

• gpt-5.5
• gpt-5.4
• gpt-5.4-pro
• gpt-5.4-mini
• gpt-5.4-nano

It also adds GPT-5.5 family detection so Ollama model_info uses gpt-5.5.context_length instead of falling back to the generic gpt.context_length key.

Fixes #419.

Why this is hardcoded for now

I checked the current Codex model info endpoint exposed by the proxy (/v1/models/gpt-5.5/info and /v1/models/gpt-5.4/info). The returned model metadata includes reasoning efforts and modalities, but does not include a context-window field such as context_window, contextLength, or context_length.

Because the upstream model info does not currently provide context length, the Ollama bridge has no dynamic source to sync from. This PR therefore keeps the existing override-table approach and updates that table to the current 400k effective context size for the GPT-5.5 / GPT-5.4 family.

Root cause

gpt-5.5 was not present in CONTEXT_WINDOW_OVERRIDES, so it fell back to 131072. GPT-5.4-family entries were still set to 272000.

Validation

• npx vitest run tests/unit/ollama/bridge.test.ts
• npx vitest run tests/unit/models/model-store.test.ts tests/unit/ollama/bridge.test.ts
• npx vitest run tests/unit
• npm run build