A 35KB binary that knows everything your team needs to know. Offline. Instant. Auditable. No servers, no subscriptions, no pages last edited in 2019.
The expense policy page hasn't been touched since 2019. The person who knew the onboarding process left in 2023. The answers live in someone's head, not on any page.
Confluence, Notion, SharePoint — they're fine tools. But they depend on humans to stay current, and humans get busy and move on.
You're paying new hires to ask questions that have been answered a hundred times before. You're holding meetings to re-explain decisions that should be documented. You're making the same mistakes because the lessons from the last incident are buried somewhere no one looks.
A Living Document Baby is your team's knowledge compiled into a native binary using the NovaGlyph Kit. Policies, procedures, fault codes, contact routing, decision history — written as FACT entries and compiled to a file you can drop on any machine.
It doesn't hallucinate. It doesn't require internet. Every answer traces back to the exact FACT line that produced it. If the answer changes, you edit one line and recompile in under 30 seconds.
Knowledge that's structural can't quietly decay — it either compiles or it doesn't. That's the guarantee a wiki page can never give you.
New hires query the binary instead of interrupting senior staff. Technicians on the floor get fault code lookups offline. Clinical staff retrieve protocols without needing network access.
Institutional knowledge stops walking out the door when people leave. It stays compiled.
When something changes — policy, personnel, process — one person edits one file, runs one compile command, and posts the new binary to Slack. The entire organisation is updated in minutes.
Every response is retrieved from your compiled FACT table — not generated, not guessed. If the fact isn't there, the Baby says so.
No servers to configure. No databases to host. No API keys to manage. Just a text file, a compiler, and a binary.
Ask two or three recent joiners what they couldn't find documented. Pull from onboarding retros. Write the raw questions down — you'll refine them in the next step.
Each question becomes one or more lines in a .light file: FACT: expense_limit IS 200_per_day. Short, direct, versioned in git. One discrete piece of information per line.
Run lightc company-knowledge.light -o company.baby. The compiler reports syntax errors immediately. Test your 20 questions, iterate until they all return correct answers.
Drop the 35KB file in Slack, your intranet, or email it. No runtime, no dependencies, no installation. Works offline on any machine. When something changes, edit one line and post a new binary.
A Baby AI fact table is not a substitute for genuine documentation of complex systems. If your architecture requires diagrams and decision trees to explain, this is the wrong tool — write real docs.
It does not replace legal review, medical judgment, or any process where reasoning about a novel situation is required. The Baby retrieves facts. It does not reason about facts it was not given. That's the design, not a limitation.
If your most experienced person left tomorrow, what would the next person not know? That's what you compile.
Precedent lookups, fee schedules, matter-type classification, client intake policy. Associates search less. Partners answer fewer repeated questions.
Protocol references, procedure codes, referral criteria. Clinical staff retrieve answers offline — including in environments without reliable network access.
Fault codes, maintenance intervals, parts substitution tables, safety procedures. Technicians on the floor query without consulting the manual or calling the office.
Onboarding, decision history, recurring process steps, escalation paths, contact routing. Anything that lives in someone's head and shouldn't.
The NovaGlyph Kit includes everything needed to build, compile, and deploy a Living Document Baby for your organisation.
Get the NovaGlyph Kit, follow the setup guide, and have your first Baby answering real questions within two hours.
We model the FACT table from your real knowledge, compile, test against your questions, and deliver a production-ready binary.