Q
Source-grounded AI for content libraries. Q ingests the video, audio, and documents a creator or company has already produced, then answers questions from that material — and cites the exact moment it came from.
The problem it solves is repetition. People with deep back catalogues answer the same questions endlessly, while the answer already exists somewhere in an old episode, talk, or document. Q makes that archive answerable on the owner’s own domain and under their own brand, with every answer carrying a citation back to the source it came from.
How it works
- Ingest. Long-form video, podcasts, and documents are transcribed and segmented into searchable passages.
- Retrieve. Semantic search runs over those passages. Embeddings are computed on Substrate’s own infrastructure using local models rather than a third-party embedding API.
- Answer with receipts. Responses are generated from retrieved passages and linked back to the source, so a reader can jump to the original moment and check it.
- Serve under your brand. Each tenant gets an isolated, branded surface with domain routing and subscriber authentication.
The answer layer routes through whichever model a tenant selects, currently defaulting to Anthropic’s Claude family. How content is handled for training or retention depends on that provider, so Q does not make a blanket promise on its behalf.
Status
Q runs in production today. A flagship long-form video tenant is live, with a corpus of more than 1,300 indexed sources. Onboarding is currently operator-led rather than self-serve: there is no public signup and no published price list. Early access starts with a conversation about library size and audience model.
At a glance
- Type
- Hosted, multi-tenant web platform
- Source
- Proprietary — the only closed-source product in the portfolio
- Built by
- Substrate Systems OÜ, founded by Hugo Ander Kivi
- Availability
- Early access, by conversation — no self-serve signup
- Product site
- useq.ai ↗