Why building it yourself is slow
The palette is the easy part. What takes a quarter is: tool calling with retries, OAuth for every connector, consent UI, audit logs, per-user cost caps, billing meters, and the dozen edge cases agents hit in production. Command+K ships those as one drop-in.
What you ship day one
- Themed ⌘K widget embedded in your app.
- One MCP server connected, tools exposed to end users.
- OAuth consent + audit log working out of the box.
- Per-user cost cap enforced.
What you ship week one
- Three to five high-value actions tied to your product's core jobs.
- Billing meter forwarding usage to your Stripe.
- End-user marketplace where users connect their own tools.
Buy the boring parts
Auth, consent, audit, billing meters — boring and unforgiving. Buying these back to your team frees a quarter to spend on the parts of your product that are actually differentiated.
FAQ
- How fast is "faster"?
- Teams typically ship their first AI action in a day and a useful palette in a week — versus a quarter to build the same thing in-house.
- What's the trade-off?
- Less control over the lowest-level UI. You get a themed widget, not a blank canvas. For 90% of teams that's the right trade.
- Can I keep my own models?
- You pick which managed models the assistant routes to, from the published catalog. Command+K itself does not support BYOK — every call goes through our managed AI Gateway so cost reporting stays exact. The action layer doesn't care which model you pick.