Connections

LiveChat AI model Back office Telegram

The AI the bot asks to pick each reply. Any OpenAI-compatible endpoint; it is called from our server, and the key never reaches a browser.

Provider

Each is tested with this bot: choose yours, paste its API key, then test.

API key Never shown again once saved. Leave the dots to keep the saved key for this endpoint.

Advanced settings

Endpoint Model id

Thinking step Skip it only on DeepSeek, Z.AI, or an endpoint that accepts enable_thinking. Response format Time limit per call (ms) Prompt cache Latest messages it reads At most this many characters of them

Longer chats keep only the latest part. Lower costs less; 40 and 12000 suit most.

Warn when credit falls under Only for a provider that reports its balance. Empty: no warning.

Backup model

Answers when the main model fails: down, out of credit, refusing its key or answering badly. Optional.

Provider

Connections

Backup model

Status

Credit left

Tokens used