The AI assistant
Snowtrol's AI assistant: real model calls via any OpenAI-compatible backend, grounded on your org's actual infrastructure with RAG over prior Q&A, plus image attachments for screenshots and diagrams.
The assistant overlay makes real model calls — there are no canned answers. The backend is any OpenAI-compatible endpoint by configuration (self-hosted Ollama or vLLM, or a hosted endpoint such as Gemini's OpenAI-compat API); when none is configured the assistant says so honestly rather than fabricating.
Grounded on your real infrastructure
Each question is answered with a system prompt built from a compact, read-only, org-scoped summary of your actual estate — resource counts by provider, connected accounts and their status, whether cost is ingested. No cross-org data and no secrets ever reach the model. On top of that, RAG retrieval finds your organization's most similar prior questions & answers (an in-database vector store, strictly org-scoped) and grounds the new answer on them; every turn's Q&A is stored back for future retrieval. If embeddings are unavailable the assistant still answers, just without retrieval.
Image attachments
Attach up to 4 images per question (PNG, JPEG, WebP or GIF, ≤5MB each) with the 📎 button — a screenshot, an architecture diagram, an error dialog. They are validated server-side and forwarded to the model as multimodal content, and they are transient: sent with the turn, never stored. A turn may be image-only.
CC_AI_BASE_URL + CC_AI_MODEL (+ an optional encrypted CC_AI_API_KEY for hosted endpoints) select the backend; the compose stack can run an internal-only Ollama service (opt-in profile). Failure modes are honest: not configured → 503, backend unreachable → 502 — never an invented answer. Requests are per-user rate-limited and audited (assistant.query). SSE streaming is a noted follow-up.