One endpoint,
every model.

Migration

From OpenAI

Change base_url to https://tare.jamerly.ai/v1 and the key to your ak-…. For a client that only sends standard fields, that is the whole migration.

From another aggregator

The endpoint is OpenAI-shaped, ids start with gen-, and unrecognised request fields are forwarded to the upstream rather than dropped, so most clients need only the base URL and key changed.

Two behaviours differ and should be checked before cutting over:

• Model names are yours to define. On the Routing page you decide what GLM 5.1 points at; nothing forces you to use the provider's naming.
• Cost on your own channels depends on the prices you enter. When the upstream reports its own cost, that value is used. When it does not — most BYOK channels — cost is computed from the cost prices you record.

The side effect of passing unknown fields through

Passing unknown request fields through is deliberate, and it is also the most common source of surprise: a parameter the upstream rejects returns the upstream’s error, not Tare’s. That is the trade-off that lets cache_control, reasoning, plugins and any parameter a provider adds later work on day one, with no release on the Tare side.

A sensible order

1. One key, one model, one call — confirm it works end to end.
2. Add your own channel and record its cost prices before real traffic arrives. Cost cannot be reconstructed later as well as it can be recorded now.
3. Route one model to your channel, leaving the platform route behind it.
4. Move the rest.
5. Set up alerts before you need them.