AI chatbotsthat liveon yourwebsites

Never. Your documents, your chats, and your knowledge base are used only to answer your visitors' questions — they're never used to train Ashh's models, your model provider's models, or anyone else's. Data is encrypted at rest, scoped to your account, and we have no read access to chat content beyond what's required to debug a billing-tier issue you raise with us.

On dedicated GPUs we operate (no shared OpenAI/Anthropic backend) — your default model is one of 10+ open-weight models we host (Qwen, Gemma, GPT-OSS, etc.). If you want a frontier model like Claude Sonnet 4.6 or GPT-4o, you can paste your own API key in Settings and pick that model per-bot. Cloud calls are opt-in and per-bot; the default is always our hosted open-weight models.

Per-token pricing surprises hurt — one viral page or a chatty visitor can blow up your bill in a day. Ashh charges a predictable monthly rate based on your tier (number of bots, sites, messages/month). When you exceed the tier, we email you and ask before upgrading — never a silent overage. Your visitors can chat as much as they want without you watching a meter.

When you upload a doc or sync a Drive folder, Ashh chunks the text into ~500-token windows, embeds each chunk with nomic-embed-text (768-dim) on our GPUs, and stores them in pgvector with HNSW indexing for fast cosine similarity search. At chat time, we embed the user query, retrieve top-k relevant chunks, and feed only those into the prompt. No prompt-stuffing, no token waste.

After creating a bot, copy a one-line <script src="…" defer></script> tag from the dashboard and paste it into your HTML. The widget renders inside a closed Shadow DOM, so your site's CSS can't leak in and the widget's CSS can't leak out. Theme, colors, position, and welcome message are all configurable.

Soon. We're shipping a browser extension, desktop app (Mac/Windows/Linux), and mobile apps so your team can talk to their bot from any tool they use, not just from a webpage. The same bot, with the same documents and same memory, accessible everywhere your team works. The public API is also live so you can build your own integrations.

Most competitors charge per-message, lock you into a single LLM provider, and silently send your customers' chats to OpenAI. Ashh has flat-rate pricing, runs your bot on open-weight models by default (with Claude / GPT-4o available via your own key), and treats your data as yours — never used to train models, never sold, never analyzed for competitive intel. Built for teams that need to pass a procurement review.