Cursor + NeoSmith

Cursor is a VS Code fork with strong native BYOM support. You have two paths:

Path	What	Endpoint	Needs Cursor Pro?	Notes
A	Claude Code extension	`router.neosmith.ai`	No	Full agentic features, scriptable via CLI
B	Cursor native BYOM	`router.neosmith.ai/v1`	Yes (Pro/Ultra)	Chat + inline edits, UI-only

Path B — Cursor native BYOM (requires Cursor Pro/Ultra)

Important: Cursor’s native BYOK cannot be configured by writing to settings.json. Keys like cursor.models.openai.* are silently ignored — Cursor stores the custom OpenAI endpoint, API key, and model list in an encrypted, server-synced store that is only editable through the UI. Custom OpenAI endpoints also require an active Cursor Pro or Ultra subscription (not available on the free tier).

Configure it in the UI:

Settings (Ctrl+,) → Models
  → scroll to "OpenAI API Key" and toggle it on
  → paste your key: sk-plus-yourname-xxxxxx
  → enable "Override OpenAI Base URL" → https://router.neosmith.ai/v1
  → "Add model": neosmith.intelligent-pro  (then Verify)
     optionally also: neosmith.intelligent-basic, neosmith.intelligent-lite
  → pick the model in the chat model picker

The CLI prints these exact values for you:

neosmith cursor on

Note: Cursor’s Tab autocomplete may still use Cursor’s own backend regardless of this setting. Chat and inline edits (Ctrl+K) route to NeoSmith.

Which to use

Want full agentic plan-mode + MCP, scriptable setup, no Pro license? → Path A (Claude Code)
Have Cursor Pro and want Cursor’s native chat/edit on NeoSmith? → Path B
Both run side-by-side without conflict.

Troubleshooting

Wrote cursor.models.* to settings.json and nothing changed: expected — Cursor ignores those keys. Use the Settings → Models UI (Path B) or the Claude Code extension (Path A).
No “OpenAI API Key” / “Override Base URL” option in Settings → Models: you’re on the free tier. Native BYOM needs Cursor Pro/Ultra — use Path A instead.
Path B verify fails: Base URL must be exactly https://router.neosmith.ai/v1.
400 Unknown model: Use neosmith.intelligent-pro, not a gpt-* name.
More: reference/troubleshooting.md