Skip to content
Aelsha Innovations

Free tool · Software Architecture Discovery, Grounded in Source

Understand an unfamiliar codebase before you change it.

Free to use

AELSHA™ NEHEMVIX helps you understand an unfamiliar codebase before you change it. You paste a Git repository link and bring your own AI model key. It reads the code at one fixed commit and writes guides for the readers you choose, from beginners to senior engineers and architects, checking every file path, name and citation against the real code. It is free to use, in the browser or from the command line: your AI provider bills model use on your own key.

Open Nehemvixnehemvix.aelsha.com

A repository is fetched at a pinned commit, passes through the pipeline stages, and produces guides for the readers chosen.

The stages, in order ↓

What Nehemvix does

It turns one analysis of a repository into guides for the readers you choose, and checks them against the code before you see them. You control the cost and keep your own model key.

  • Guides from one analysis

    Guides for beginners, senior engineers and architects come from a single pass over the code, with architecture diagrams drawn from its real structure. Each reads at three depths: overview, guide and deep dive.

  • Checked against the code

    Invented names and paths are detected, citations resolve to exact lines, and a weak section gets one repair attempt. Each guide is scored 0 to 100, and no AI model acts as judge.

  • Careful with the repository you point it at

    The fetch is pinned to a commit. Secrets found in the repository are replaced before any model call, and text in the repository that is addressed to AI tools is flagged.

  • Cost and keys stay with you

    You see an estimate and approve it before anything is spent, with a spending cap and a ledger of calls. Your model key is held in memory only: it is sent with each run and never written to a database, log or file.

How it works

Four steps from a link to a checked guide. Nothing is spent until you approve the estimate.

See the structure. Connect the relationships. Check the evidence.

  1. 01

    Paste a link, bring a key

    Give it a Git repository link and your own AI model key. The code is fetched at one fixed commit.

  2. 02

    Approve the estimate

    Nehemvix estimates the cost and pauses. You approve it, or stop, before any model is paid for.

  3. 03

    It reads, writes and checks

    It maps the code, writes each guide, then checks every file path, function name and citation against the real code.

  4. 04

    Read, export, ask

    Read the guides, export them as Markdown, PDF, Word, PowerPoint, HTML or draw.io, and ask questions about the repository with the same name checks.

Tested with OpenAI, Anthropic Claude, and Google Gemini and Vertex AI. Amazon Bedrock, Microsoft Foundry and OpenAI-compatible endpoints can also be set up, always with your own key.

How it is built

Nehemvix is an AI agent built as a structured pipeline, not a free-roaming one. The model that writes has no tools, and the checks run in code, so a guide cannot grade itself.

The stages, in order

  1. Repository: you paste a Git link.
  2. Pinned fetch: the code is read at one fixed commit.
  3. Ingest checks: secrets are redacted and text addressed to AI tools is flagged.
  4. Estimate and approval: you see the cost and a person approves it.
  5. Map and reduce: the code is mapped, then reduced to what matters.
  6. Write: each guide is drafted.
  7. Evaluate: names, paths and citations are checked against the code, and each guide is scored.
  8. Guides: for the readers you choose, from beginner to architect.

Every claim can point to a line

A claim in a guide carries a citation that resolves to an exact line of the code at the pinned commit. If it does not resolve, it is flagged or repaired. The format, shown with a placeholder path:

Claim in a guide: “Requests are authenticated before they reach the router.”path/to/file.py:120

Illustrative format, not a real repository.

Use it where you work

The web app, the command line and AI assistants share one engine, and each uses your own model key.

  • In the browser

    Sign in at nehemvix.aelsha.com, paste a link, approve the estimate, and read, question and export the guide there.

  • From the command line

    Install it from PyPI and run the same pipeline on your own machine, or in CI with a minimum grounding score.

    pipx install aelsha-nehemvix

  • From an AI assistant

    Give Claude or another assistant read-only access to list, read and search your guides, over MCP or A2A. Claude connects as a custom connector when you sign in; other assistants sign in the same way where they support it, or use a token.

Planning AI agents for your own systems?

Aelsha Innovations’ consulting starts with a fixed-scope readiness assessment.

Explore agentic-AI consulting