Simple, honest pricing

No. You bring your own API key from your chosen provider (OpenAI, Anthropic, Google Gemini, etc.). All AI costs go directly to your provider account. Ottoch never sits between you and your provider's billing.

No. Ottoch runs entirely on your own server. There are no analytics, no telemetry, and no call-home. The only outbound network calls from your instance are to your chosen AI provider's API endpoint — and those calls contain only your schema and your question, never your row data.

The app continues to work without interruption. On the next monthly query count reset, it downgrades to free tier limits (50 queries, 1 connection). Your data, sessions, and configuration are untouched. Renew at any time to restore enterprise limits.

Yes. License validation is offline — the license key is checked locally. The only internet dependency is your AI provider's API. If you're running a local model via an OpenAI-compatible endpoint, Ottoch works in a fully air-gapped environment.

Yes. Pick Custom (OpenAI-compatible) as the provider in LLM Configuration and point the Base URL at your local server's OpenAI-compatible endpoint — for Ollama that's /v1/chat/completions, not its native /api/chat path, since the two use different request/response formats. A bare hostname works fine too (e.g. a Docker container or service name like http://ollama:11434/v1/chat/completions) — it doesn't need to be a real internet domain. Use the Fetch models button after entering the URL to pick from whatever models your local server actually has loaded, rather than typing the exact name by hand.

The most common cause is a table that's needed to answer the question isn't activated for that connection — Ottoch only ever sees the schema of tables you've explicitly turned on under Manage tables. A question like "top-selling products" needs whichever table actually links orders to products (an order-items / line-items table), not just the products and orders tables on their own — activate every table the question touches, including join tables, and try again. If every relevant table is already active and it still can't answer, a smaller local model may simply be struggling with that particular query; a larger model (GPT-4o, Claude, Gemini) tends to handle multi-table joins more reliably.

Email us at [email protected] or use the contact form. Enterprise licenses include priority response — we aim to reply within one business day. Free tier support is best-effort.