Cabi.Chat Ignite
Cabi.Chat Ignite is the graph's conversation engine. Radar finds; Ignite speaks. Every oracle reply, widget answer, first contact and LLM translation is one Ignite run: your instructions, the live conversation and the tools you granted. No script, no canned replies.
What runs on it
- Telegram account oracles and bot oracles. Cabi.Chat Oracles — Telegram account oracle Cabi.Chat Oracles — Telegram bot oracle
- The website widget. Cabi.Chat Oracles — Website widget
- The reply to a public request found by Radar, written by the same agenda that continues the chat. link Cabi.Chat Campaigns
- Card translation with your own translation agenda. Cabi.Chat Graph — Languages and translation
Every run follows an agenda: Cabi.Chat Ignite — Agendas
The turn. Messages a person sends in quick succession are read and answered together, as one turn. Inside a turn the model works step by step: it reads, looks things up, calls tools and sends its answer through the surface's send tool. A turn may also end in silence when nothing should be said. What it can call: Cabi.Chat Ignite — Tools, endpoints and secrets Which model runs it: Cabi.Chat Ignite — Models and AI usage
Notes about the person. Ignite keeps short notes on each conversation: name, city, what the person wants, what was promised. They are read on every next turn, belong to that one conversation and are used nowhere else. A person can ask for them to be deleted, and the oracle can clear them.