AI
AI in RolChat is a layer over every module, not a module of its own, and it runs on your provider keys. RolChat provides the adapters, the interface, orchestration, caching and cost transparency; you pay the vendor directly.
Providers
Section titled “Providers”Nine adapters are in the catalogue: OpenAI, Anthropic, Google Gemini, xAI Grok, DeepSeek, Alibaba Qwen, MiniMax, ElevenLabs and Deepgram, covering chat, embeddings, speech-to-text, text-to-speech, vision and voice agents. The model list (94 models) is refreshed weekly. Connect a provider in Settings → AI routing → Provider catalogue: paste the key, RolChat validates it, and you choose which model handles which capability. One key per vendor; several vendors can be active at once.
Keys are stored encrypted, written only by the plugin service, and never leave the platform except in calls to the vendor.
What AI does
Section titled “What AI does”| Where | Capability |
|---|---|
| Conversations | Suggested replies, rewrite (shorter, friendlier, formal), thread summary, live translation per message, language and sentiment detection |
| The AI agent | Answers new conversations first when auto-reply is on, using your knowledge base and instructions; hands over to a person when asked, when unsure, or when your rules say so |
| Onboarding | The AI concierge sets up the account from a conversation and escalates to RolChat support when stuck |
| Knowledge base | Gap detection drafts articles from unanswered questions; article translation |
| Phone | Transcription, call summaries, call scoring, an IVR auto-attendant node, a voice agent |
| CRM | Health score, data-quality suggestions, identity graph embeddings |
| Automation | Classify, summarise and suggest as rule actions |
| Privacy | Policy generation and per-paragraph rewriting |
Teaching the AI agent
Section titled “Teaching the AI agent”Four sources feed the agent’s context: knowledge base articles, uploaded documents, approved past answers, and instructions you write. Managers approve what enters the context; the agent says what it does not know rather than inventing, and never answers on billing or legal matters unless you allow it.
Budgets and usage
Section titled “Budgets and usage”Every AI call emits a usage record with tokens and estimated cost. Settings → AI routing → Usage shows spend per capability and per vendor, and a per-account monthly cap stops calls when reached. The Reports AI dashboard shows outcomes: conversations handled and resolved by AI, escalations and their reasons, AI CSAT.
If your vendor refuses a call for billing reasons the product says so plainly (error code ai_vendor_out_of_funds) instead of reporting an internal failure.
Without a key
Section titled “Without a key”Everything still works, without the AI parts: no suggested replies, no auto-reply, the IVR falls back to key presses, knowledge base search is keyword-based, the AI dashboard and AI rule actions are hidden. Connecting a key later switches them on; nothing has to be reconfigured.