Read and use your guide
Start with the overview, go deeper when you need to, and check any claim against the exact lines of code.
A guide at a glance
- Grounding score, 0 to 100, with a word beside it: Strong, Fair or Check. About the score
- The verdict: All found when every name, path and citation is in the code; otherwise how many names weren't found, with Review them. About missing names
- Depth: Overview, Guide or Deep dive. About depth levels
- Show evidence: each citation becomes a chip that opens the exact lines. About evidence
- Ask a question about the code, beside the guide. About questions
- Export to PDF, Word, PowerPoint, Markdown, HTML or draw.io. About exports
The drawing uses the real top of a guide Nehemvix wrote for pallets/itsdangerous. Read a whole guide written at the same commit.
Depth and audiences
- Overview, Guide, Deep dive change how much of each section you see; a line under the switch says which: Overview is the quick start and a short summary of each part, Guide is every part in full without the module and file reference, and Deep dive is everything. Your place is remembered when you come back.
- If you made more than one guide (beginner, senior, architect), switch between them at the top.
Evidence and the score
- Tick Show evidence: every citation becomes a chip that opens the exact lines at the commit that was read, and paragraphs that name code without citing it are marked needs a source.
- The grounding score (0–100) measures how well the guide sticks to what is really in the code: real names and paths, resolvable citations, coverage, a valid diagram. It does not prove every explanation is right. For anything important, follow the citation. How we check. Beside the score a word sums it up: Strong (90 and up), Fair (70 to 89) or Check (below 70).
- Under the title, one sentence says whether every name was found in the code, or how many weren't (for example 2 names in this guide weren't found in the code); Review them lists those names.
- With Show evidence on, each section heading says how many source files it cites, or no citations.
Readiness
Readiness lists signals found in the code, each with the files that show it: CI, tests, packaging, health checks, configuration, logging, dependency pinning and licences, and credentials in code (file and line only, never the value). For AI and agent applications it adds observability, evaluation, guardrails, cost control and human approval, with an indicative observability stage from 1 (untraced) to 4 (cost-controlled with oversight). For a service it also shows Cloud readiness and a Cloud Run or GKE fit (Ready or Needs changes, with the changes). Each line is Pass (✓), Worth reviewing (▲) or Missing (✕): whether something exists in the code, not whether it works. These are automated signals, not a security audit.
How it was made
How it was made says in plain words what ran, what it cost and what was checked, with the model calls, tokens, cost, file summaries reused and the grounding score, and the price used with its source. Technical details holds the rest: the pipeline as it ran, calls, tokens and cost by stage, the prompt-cache rate, sections repaired, the cost formula, and FinOps and token optimisation with each saving and this guide's number.
Ask about this code
Ask questions beside the guide, for example Where does the program start? Answers cite the code and say which names they couldn't find. Each answer says which model wrote it and what it cost.
Questions can use any provider and model, not only the one that wrote the guide: open Model for answers, choose the provider and model, and enter its key. If that model hasn't seen this repository yet, tick I agree to send excerpts of this repository to … first; Ask stays disabled until you do. Rate any answer Helpful or Not helpful: the vote is stored with that answer, one per answer (a second click replaces it), costs nothing and is not sent to your AI provider. It tells us how well Ask works; answers never become test data. If your key is no longer in this tab, the panel asks for your provider's details again.
Rate the guide
Under the guide, answer Was this guide accurate? A No asks what was wrong. Ratings on guides of public repositories help us test Nehemvix; private code never becomes test data.
Export
Press Export, choose a format and a style, optionally your Main colour, Accent colour and Company name (optional) for the PDF, Word and PowerPoint styling, then Download:
| Format | Styles |
|---|---|
| Report: a cover with the audience and grounding score, numbered sections, contents, page numbers, numbered figures, sources after each section, tagged for screen readers | |
| Word | Editable report with a contents page that updates itself |
| PowerPoint | Briefing deck (one slide per section, details in the speaker notes) or Detailed deck |
| Markdown | Front matter for documentation sites, links inline |
| HTML | One self-contained page with light and dark modes that also reads well on a phone |
| Diagrams | Every diagram as an editable draw.io file |
PDF, Word, PowerPoint and HTML exports carry the AELSHA™ mark in the footer with "Made with Nehemvix by AELSHA™ · Aelsha Innovations"; with a company name, your name replaces the attribution and the small mark stays.
Delete a guide
Press Delete this guide below the guide, type DELETE and confirm. The guide, its exports, questions and saved progress are removed. See Your data.