AI Instruments Configuration

This section describes the AI instruments of the Citeck platform: configurable LLM-based agents, the transcription microservice, and the meeting recording module.

Custom AI agents are configurable entities with a system prompt, a set of tools, and a selected LLM provider. Agents operate in two modes: stateless (single invocation from BPMN processes) and stateful (multi-turn dialogue in the AI Assistant UI). Administrators create agents via Admin > AI Agents; developers can connect custom tools by implementing the CiteckAiTool interface.

Audio transcription and diarization microservice — runs as a sidecar to the citeck-ai service. Accepts an audio file (WebM, WAV), converts it to WAV 16 kHz via ffmpeg, transcribes it in 25-second chunks using the GigaAM-v3 model, and, when a HuggingFace token is provided, identifies speakers via pyannote.audio. Available on port 8090.

Meeting recording module captures audio from a browser tab and microphone via the Chrome extension, streams it over WebSocket, transcribes it using citeck-stt-sidecar, and automatically generates a summary via LLM. The result is saved as an ECOS activity of type meeting-activity or call-activity. The module is available starting from version 1.11.0.