One endpoint,
every model.
Tare is an OpenAI-compatible gateway in front of every model you use — ours and the ones you bring yourself. One base URL, one key, and every token accounted for.
Alerts
Console → Alerts. Rules watch spend, error rates and channel failures; channels decide where the notice goes.
Delivery channels
| Channel | Target |
|---|---|
| Webhook | Any URL; Tare sends a JSON body by POST |
| Feishu / Lark | A group bot webhook, with optional signature verification |
Saving replaces every channel; anything left blank is removed. Existing targets are shown masked and cannot be edited in place, because a webhook URL is a credential and is never sent back.
Feishu signature verification
Fill in the signing secret if the bot has signature verification switched on, and leave it empty if it is off.
[!WARNING] Getting this wrong is silent. A bot with verification on rejects an unsigned message, and the reason stays on Feishu's side — the platform sees an error code and you see no alerts. Nor can the setting be detected: which setting the bot has, because a bot with verification off also accepts a signed message. The setting can only be confirmed on your side.
Feishu also only accepts a timestamp within an hour of its own clock, so a badly drifted container looks exactly like a wrong secret.
There is no platform fallback channel
With no channel configured, an alert is recorded as undeliverable and stops there; it is not sent to the platform instead. Routing customer spend alerts into the platform's operations chat would be a data leak, not a safety net.