Grounded AI products for the web. Cited answers, not chatbot theatre.
We build AI-native products that answer real questions with real sources — for podcasts, sermons, scripture, and the wider web. Every claim links back to the page it came from. When the content doesn't support the question, the product says so.
What we believe
Answers over agents. Sources over swagger.
Search engines return hits. Chatbots invent things. We build a third thing: narrative answers where every sentence is tied to a real page in real content — with the humility to refuse when the source isn't there.
Every claim linked
Inline citations, source chips, and click-through to the exact page and anchor. If we can't cite it, we don't say it.
Refuse when ungrounded
"I don't see that in your content" is a feature, not a bug. A single confident hallucination reads worse than ten honest "I don't know"s.
Self-serve by default
Sign up, add a source, paste two lines of HTML. No sales calls, no procurement, no live-chat handoff. Credit card and go.
Live products
Answer engines you can use today.
Three shipped products, one platform underneath. All three use the same idea — narrative, cited answers over a specific corpus — applied to a specific audience.
Podcasts
Ask the archive anything.
An embeddable widget that turns any podcast's back catalogue into a searchable, cited answer engine. Ask, search, and jump to the exact episode moment.
podsaid.com
Churches
Sermons your congregation can search.
Every message a pastor has ever preached, indexed and answerable. Members ask real questions and get a paragraph pointing back to the sermon that answered it.
sermons.podsaid.com
Scripture
Every answer, chapter and verse.
A Bible AI that reasons across translations and returns cited passages — never a paraphrase without a reference. Study, discipleship, and honest disagreement, grounded in text.
biblecorral.comIn development
Bringing the same discipline to the wider web.
Next
Perplexity for your website.
An embeddable answer engine that indexes any site — docs, product pages, long-tail archives — and serves narrative, cited answers to visitors. Same rails as PodSaid, aimed at the whole open web.
How we build
Repo-Driven Development.
Repo-Driven Development is human-in-the-loop AI engineering. A person sets intent, reviews every diff, and holds the merge button. The AI does the drafting and toil — inside a repo shaped to make that collaboration safe and fast.
That's the unlock: a small team ships production AI products at a pace that used to require a team ten times its size, because every repo is a first-class collaborator — structured layout, GitLab as ALM, AI-assisted development, and continuous scheduling from day one.
- Human in the loop, by design. The AI drafts and reads; the human sets intent, reviews the diff, and hits merge. No unattended agent commits, ever.
- Structured repos with landing pages, milestones, and CI schedules by default
- ALM and issues from the CLI —
glab, not clicking - Automated sync so every repo stays coherent as the fleet grows
AI safety
We take AI safety seriously.
Not as a banner on a website. As the thing that decides what we ship, what we refuse to ship, and how loudly we are willing to disagree with our own industry.
The honest risk in the products we build is not a rogue superintelligence. It is a confident machine quietly telling someone something untrue, in a voice that sounds like it checked. That is the failure we engineer against, and it is why every answer we ship carries its sources and every ungrounded question gets a refusal instead of a guess.
- A human holds the merge button. Our own development is human-in-the-loop by design — the AI drafts, a person reviews the diff and ships it. No unattended agent commits, in our products or in our repos.
- Refusal is a feature. A product that says "I don't see that in your content" is safer than one that is never wrong out loud.
- Citations are not decoration. If a claim can't be traced to a page, it doesn't get made.
We also put money and time into the unglamorous half of this: helping people think clearly about AI instead of panicking about it on a monthly cycle. That's why we sponsor Keep Calm and Prompt On.
We sponsor
Keep Calm and Prompt On
A monthly magazine that de-hypes the AI industry — one careful read of what was actually claimed, what was true, and what it means for the people using these tools. No doom, no engagement bait.
keepcalmandprompton.orgSay hello
Want to embed grounded answers in your own product?
We're building for podcasters, pastors, publishers, docs teams, and anyone else with content that visitors deserve real answers about. Send us a note.
info@middlewaysoftware.com