Run Novix on your own inference account

How Novix gets involved

Inference settings on the Pro plan.

One screen in the dashboard: pick a provider, paste a key, save. The key is probed before it is stored, so a wrong one fails at the settings page rather than on your next ticket.

How the agent handles it

Six providers, two API shapes
Anthropic and Kimi speak the Anthropic shape and keep every engine feature. OpenAI, Google Gemini, DeepSeek and OpenRouter go through an adapter that loses prompt caching and the effort dial, so a run there costs you more.
The key is tested before it is trusted
Saving a key probes it, and that probe goes through the same guard every connector uses: the host is resolved once and the vetted address pinned, re-checked at every redirect.
A refused key never falls back to ours
The stage fails and names which provider refused it, and your admins are mailed once. Falling back would move the token bill onto Novix at exactly the moment your configuration broke.
Whose key paid is a fact about the run
The tokens are spent during the run and the approval that bills happens later, so the meter reads the usage ledger for that task rather than whatever the settings page says at the time.
What that does to the bill
On your own key, approved fixes are unlimited and nothing is metered. On Novix’s key each approved fix is $1.49, and on Pro, where bringing a key is the alternative, there is no ceiling on that.

What your team gets

The inference bill on your account, under your own provider agreement
Unlimited approved fixes, with nothing metered
A ledger row per model call recording the model, the lane and whose key paid

This is not a claim that Novix is model agnostic

The nine stages, the effort dials and the token ceilings are sized against Claude. Bringing a key is a cost, compliance and procurement choice, and it changes whose account pays rather than which model does the work.

Providers you can point it at

Anthropic
Kimi
OpenAI
Google Gemini
DeepSeek
OpenRouter

The first two keep every engine feature. The rest go through the chat-completions adapter and lose prompt caching and the effort dial.

Try it on a real customer problem.

Connect the place reports or product signals already arrive. The same agent takes it from evidence to a tested pull request.

Start free Chat with the founder

How the agent handles it

Six providers, two API shapes

Anthropic and Kimi speak the Anthropic shape and keep every engine feature. OpenAI, Google Gemini, DeepSeek and OpenRouter go through an adapter that loses prompt caching and the effort dial, so a run there costs you more.

The key is tested before it is trusted

Saving a key probes it, and that probe goes through the same guard every connector uses: the host is resolved once and the vetted address pinned, re-checked at every redirect.

A refused key never falls back to ours

The stage fails and names which provider refused it, and your admins are mailed once. Falling back would move the token bill onto Novix at exactly the moment your configuration broke.

Whose key paid is a fact about the run

The tokens are spent during the run and the approval that bills happens later, so the meter reads the usage ledger for that task rather than whatever the settings page says at the time.

What that does to the bill

On your own key, approved fixes are unlimited and nothing is metered. On Novix’s key each approved fix is $1.49, and on Pro, where bringing a key is the alternative, there is no ceiling on that.