AELSHA™ Nehemvix
Menu
Start a free guide

Architecture

One engine makes every guide. Run it in the browser or on your own machine, and read your guides from an AI assistant.

AELSHA™ NEHEMVIX is one Python package, nehemvix. The web app and the command line both run the same pipeline from it, with the same audiences, the same grounding checks and the same price list. AI assistants don't run the pipeline: they read the guides that the web app keeps in your account.

One engine, three ways in

One engine, three ways in Three ways to use Nehemvix. The web app, in your browser with your key in the tab's memory, sends your key with the run to Nehemvix's service, which runs the pipeline (Read, Understand, Write, Check) and keeps your guides in your account. An AI assistant connects over MCP or A2A with an access token and only reads the guides in your account. The command line runs the same pipeline on your own machine, with your key from an environment variable, and writes your guides as files in ./out, where assistants can't see them. Both the service and your machine fetch the code from your Git host and call the model you chose with your key. One package, nehemvix, gives both the same pipeline, audiences, grounding checks and price list. Web app In your browser Key in the tab's memory AI assistant MCP or A2A An access token Your key, with the run Reads only Nehemvix's service Runs the pipeline with your key Read Understand Write Check Your guides, kept in your account Fetches the code, calls the model Your Git host and the model you chose Command line, on your machine Your key from an environment variable Read Understand Write Check Your guides, as files in ./out Assistants can't see these Fetches the code, calls the model Your Git host and the model you chose One package, nehemvix The same pipeline, audiences, grounding checks and price list in both places One engine, three ways in Three ways to use Nehemvix. The web app, in your browser with your key in the tab's memory, sends your key with the run to Nehemvix's service, which runs the pipeline (Read, Understand, Write, Check) and keeps your guides in your account. An AI assistant connects over MCP or A2A with an access token and only reads the guides in your account. The command line runs the same pipeline on your own machine, with your key from an environment variable, and writes your guides as files in ./out, where assistants can't see them. Both the service and your machine fetch the code from your Git host and call the model you chose with your key. One package, nehemvix, gives both the same pipeline, audiences, grounding checks and price list. Web app In your browser Key in the tab's memory AI assistant MCP or A2A An access token Your key, with the run Reads only Nehemvix's service Runs the pipeline with your key Read Understand Write Check Your guides, kept in your account Fetches the code, calls the model Your Git host and the model you chose Command line, on your machine Your key from an environment variable Read Understand Write Check Your guides, as files in ./out Assistants can't see these Fetches the code, calls the model Your Git host and the model you chose One package, nehemvix The same pipeline, audiences, grounding checks and price list in both places

The web app

Runs on Nehemvix's service. You sign in, approve the estimate, and your guides are kept in your account.

The command line

Runs on your own machine. It talks to your Git host and your model provider directly, and writes your guides as files.

AI assistants

Read the guides in your account, with an access token you can revoke. They make no model calls and spend nothing.

Your key, your provider

Every guide is written with your own model key, by the provider you chose. Guides made on the command line stay on your machine, so an assistant can't see them.

Which one to use

Web app Command line AI assistants
Where it runs Nehemvix's service Your machine Your assistant calls Nehemvix's service
Your model key In your browser tab's memory, sent with the run In an environment variable Not needed: no model calls
Approving the cost You approve the estimate in the app --cap (default $1.00): above it, the run stops before spending Nothing is spent; prepare_guide opens the app, where you approve
Where guides are kept In your account Files in ./out; progress in ~/.nehemvix Reads the guides in your account
Ask Beside the guide nehemvix ask Your assistant searches and reads sections, and answers with its own model
Exports PDF, Word, PowerPoint, Markdown, HTML, draw.io Markdown, plus --format html,pdf,docx,pptx,drawio Sections as Markdown, with citations
Private repositories A read-only token, used for one fetch, never stored --git-token-env: one fetch, never stored Reads their guides like any of yours
Best for Trying it, then reading and asking in one place CI, scripts, and keeping guides on your machine Asking your assistant about guides you already have

The web app

How the web app works Your browser runs the app at /app, signed in with Google, and keeps your model key in the tab's memory only; you read your guide there, ask about it and export it. The key is sent with the run to Nehemvix's service on Google Cloud, which sends back progress and your guide. The service runs none of the repository's code, removes secrets before any model call, waits for you to approve the estimate, and keeps finished work when you pause and resume. Your guides and progress are kept in your account. It makes one fetch at a pinned commit from your Git host, and sends code excerpts with your key to the model you chose. Package names and versions only, never code, go to public package registries. Your browser The app at /app, signed in with Google Your key stays in this tab's memory Read your guide, ask about it, export it Sent with the run Progress and your guide Nehemvix's service On Google Cloud Runs none of the repository's code Removes secrets before any model call Waits for you to approve the estimate Pause and resume keep finished work Your guides and progress, in your account Your Git host One fetch at a pinned commit The model you chose Code excerpts, with your key Package names and versions only, never code, go to public package registries. How the web app works Your browser runs the app at /app, signed in with Google, and keeps your model key in the tab's memory only; you read your guide there, ask about it and export it. The key is sent with the run to Nehemvix's service on Google Cloud, which sends back progress and your guide. The service runs none of the repository's code, removes secrets before any model call, waits for you to approve the estimate, and keeps finished work when you pause and resume. Your guides and progress are kept in your account. It makes one fetch at a pinned commit from your Git host, and sends code excerpts with your key to the model you chose. Package names and versions only, never code, go to public package registries. Your browser The app at /app, signed in with Google Your key stays in this tab's memory Read your guide, ask about it, export it Sent with the run Progress and your guide Nehemvix's service On Google Cloud Runs none of the repository's code Removes secrets before any model call Waits for you to approve the estimate Pause and resume keep finished work Your guides and progress, in your account Your Git host One fetch at a pinned commit The model you chose Code excerpts, with your key Package names and versions only, never code, go to public package registries.

How it works, step by step

  1. Open the app and sign in with Google.
  2. Paste a repository link, choose your provider and model, and give your key. It stays in this tab's memory and is sent only with the run.
  3. Nehemvix's service fetches the repository once, at one pinned commit. It runs none of the code and removes secrets before any model call.
  4. You see the estimate and approve it. Nothing is spent before you do.
  5. Code excerpts go, with your key, only to the provider and model you agreed to.
  6. Your guide and its progress are kept in your account. If a model limit is reached or you pause, resume later: finished work is kept and never billed twice.
  7. Read the guide in the app, ask questions beside it, and export it.

Step by step with every option: Make your first guide.

The command line

How the command line works You install aelsha-nehemvix from PyPI with pipx. On your machine you run nehemvix run, resume, finish, ask or doctor, read your guides with nehemvix list, status, outline, section, search and readiness, or serve them to an AI agent with nehemvix mcp, with your key from an environment variable. Nehemvix runs the same pipeline as the web app there, and --cap is your approval: a run whose estimate is above it stops before spending. Progress and file memory stay in ~/.nehemvix; guides, the checklist and the report are written to ./out. Your machine makes one fetch at a pinned commit from your Git host and sends code excerpts with your key to the model you chose, never through nehemvix.aelsha.com. Package names and versions only, never code, go to public package registries. Exit codes for scripts and CI: 0 done, 1 stopped, 2 wrong input, 3 paused, 4 needs a person's approval (with --json), 130 Ctrl-C. From PyPI pipx install aelsha-nehemvix Your machine Your key from an environment variable nehemvix run nehemvix resume nehemvix ask nehemvix doctor nehemvix list nehemvix status nehemvix outline nehemvix section nehemvix search nehemvix readiness nehemvix mcp nehemvix finish Nehemvix The same pipeline as the web app --cap is your approval: over it, it stops before spending ~/.nehemvix Progress and file memory ./out Guides, the checklist and the report Your Git host One fetch at a pinned commit The model you chose Code excerpts, with your key Never through nehemvix.aelsha.com Package names and versions only, never code, go to public package registries. Exit codes for scripts and CI 0 done 1 stopped 2 wrong input 3 paused 4 approval 130 Ctrl-C How the command line works You install aelsha-nehemvix from PyPI with pipx. On your machine you run nehemvix run, resume, finish, ask or doctor, read your guides with nehemvix list, status, outline, section, search and readiness, or serve them to an AI agent with nehemvix mcp, with your key from an environment variable. Nehemvix runs the same pipeline as the web app there, and --cap is your approval: a run whose estimate is above it stops before spending. Progress and file memory stay in ~/.nehemvix; guides, the checklist and the report are written to ./out. Your machine makes one fetch at a pinned commit from your Git host and sends code excerpts with your key to the model you chose, never through nehemvix.aelsha.com. Package names and versions only, never code, go to public package registries. Exit codes for scripts and CI: 0 done, 1 stopped, 2 wrong input, 3 paused, 4 needs a person's approval (with --json), 130 Ctrl-C. From PyPI pipx install aelsha-nehemvix Your machine Your key from an environment variable nehemvix run nehemvix resume nehemvix ask nehemvix doctor nehemvix list nehemvix status nehemvix outline nehemvix section nehemvix search nehemvix readiness nehemvix mcp nehemvix finish Nehemvix The same pipeline as the web app --cap is your approval: over it, it stops before spending ~/.nehemvix Progress and file memory ./out Guides, the checklist and the report Your Git host One fetch at a pinned commit The model you chose Code excerpts, with your key Never through nehemvix.aelsha.com Package names and versions only, never code, go to public package registries. Exit codes for scripts and CI 0 done 1 stopped 2 wrong input 3 paused 4 approval 130 Ctrl-C

How it works, step by step

  1. Install it from PyPI with pipx install aelsha-nehemvix; nehemvix doctor checks your machine.
  2. Put your key in an environment variable: the provider's usual one, or another you name with --key-env.
  3. Run nehemvix run <repository> --provider … --model … --cap 0.50.
  4. It prints the estimate. If the upper end is within --cap, it goes on; otherwise it stops before spending anything. On the command line, --cap is your approval.
  5. Your machine fetches the repository from your Git host and calls your model provider directly. Nothing goes through Nehemvix's service.
  6. Guides, the readiness checklist and the report go to ./out. Progress and remembered file summaries stay in ~/.nehemvix.
  7. Paused or stopped? nehemvix resume <job-id> continues on any model. nehemvix ask asks a finished guide a question.
  8. The exit code tells a script or CI what happened.
  9. For your own AI agents: --json prints one result object, nehemvix list, outline, section and search read your guides, and nehemvix mcp serves them over MCP on your machine. See Use Nehemvix from AI agents.

As in the web app, dependency names and versions (never code) are looked up on public package registries for public repositories. Every command and option: Command line.

AI assistants

How AI assistants connect Your AI assistant, any MCP client such as Claude Code or Cursor, or an A2A agent, connects with an access token from the app's Settings, which expires and can be revoked at any time. MCP clients use nehemvix.aelsha.com/mcp; A2A agents use nehemvix.aelsha.com/a2a, with the agent card at /.well-known/agent-card.json. Nehemvix gives the assistant 7 read-only tools: list_guides, get_guide_status, get_guide_outline, get_section, search_guide, get_readiness and prepare_guide. They make no model calls, spend nothing and read only your own guides, kept in your account. prepare_guide returns a link to the web app, which opens a new guide with the repository filled in; you approve the cost there. Guides made on the command line stay on your machine, so assistants can't see them. Your AI assistant Any MCP client, such as Claude Code or Cursor, or an A2A agent An access token from Settings It expires; revoke it at any time MCP nehemvix.aelsha.com/mcp A2A nehemvix.aelsha.com/a2a Agent card /.well-known/agent-card.json Nehemvix 7 read-only tools list_guides get_guide_status get_guide_outline get_section search_guide get_readiness prepare_guide No model calls, nothing spent Only your own guides Your guides, in your account prepare_guide returns a link The web app Opens a new guide with the repository filled in. You approve the cost there. Guides made on the command line stay on your machine: assistants can't see them. How AI assistants connect Your AI assistant, any MCP client such as Claude Code or Cursor, or an A2A agent, connects with an access token from the app's Settings, which expires and can be revoked at any time. MCP clients use nehemvix.aelsha.com/mcp; A2A agents use nehemvix.aelsha.com/a2a, with the agent card at /.well-known/agent-card.json. Nehemvix gives the assistant 7 read-only tools: list_guides, get_guide_status, get_guide_outline, get_section, search_guide, get_readiness and prepare_guide. They make no model calls, spend nothing and read only your own guides, kept in your account. prepare_guide returns a link to the web app, which opens a new guide with the repository filled in; you approve the cost there. Guides made on the command line stay on your machine, so assistants can't see them. Your AI assistant Any MCP client, such as Claude Code or Cursor, or an A2A agent An access token from Settings It expires; revoke it at any time MCP nehemvix.aelsha.com/mcp A2A nehemvix.aelsha.com/a2a Agent card /.well-known/agent-card.json Nehemvix 7 read-only tools list_guides get_guide_status get_guide_outline get_section search_guide get_readiness prepare_guide No model calls, nothing spent Only your own guides Your guides, in your account prepare_guide returns a link The web app Opens a new guide with the repository filled in. You approve the cost there. Guides made on the command line stay on your machine: assistants can't see them.

How it works, step by step

  1. In the app, open Settings → Connect an AI assistant and create an access token. It lasts 7, 30 or 90 days, and you can revoke it at any time.
  2. Add Nehemvix to your assistant: any MCP client, such as Claude Code or Cursor, at https://nehemvix.aelsha.com/mcp, or an A2A agent at /a2a, with the agent card at /.well-known/agent-card.json.
  3. Your assistant gets seven read-only tools: list_guides, get_guide_status, get_guide_outline, get_section, search_guide, get_readiness and prepare_guide.
  4. Nehemvix answers from the guides in your account. It makes no model calls and spends nothing; your assistant writes its answer with its own model.
  5. To make a new guide, prepare_guide returns a link that opens the app with the repository filled in. You choose the model and approve the cost there.
  6. An assistant sees only your own guides. Guides made on the command line stay on your machine, so it can't see them; for those, run nehemvix mcp locally (Use Nehemvix from AI agents).

Tokens and a client example: Connect an AI assistant.

What they share

The web app and the command line run the same four stages, with the same audiences, grounding checks and price list.

How a guide is made: Read, Understand, Write, Check From left to right. Your repository link, at one commit. 1 Read: fetches the code, removes secrets and runs none of it; nothing is sent to a model yet. Then the gate: you approve the model, consent and cost. 2 Understand: each file, then folders, then the whole system. 3 Write: each section for your audiences, with real diagrams. 4 Check: looks up names, links exact lines and scores the guide. Guide ready: checked before you see it. Under Understand and Write: paused at a limit or by you? Finished work is kept and never billed twice; resume on any model. You approve the model, consent and cost Repository Your link, one commit 1 Read Fetches the code Removes secrets Runs none of it Nothing sent yet 2 Understand Each file, then folders, then the whole system 3 Write Each section for your audiences, real diagrams 4 Check Looks up names Links exact lines Scores the guide Guide ready Checked before you see it Paused at a limit or by you? Finished work is kept and never billed twice; resume on any model. How a guide is made: Read, Understand, Write, Check From top to bottom. Your repository link, at one commit. 1 Read: fetches one commit, removes secrets and runs none of the code; nothing is sent to a model yet. Then the gate: you approve the model, consent and cost. 2 Understand: each file, then folders, then the whole system. 3 Write: each section for your audiences, with real diagrams. 4 Check: looks up names, links exact lines and scores the guide. Guide ready: checked before you see it. Paused at a model limit, or by you? Finished work is kept and never billed twice; resume on any model. Repository Your link, one commit 1 Read Fetches one commit. Removes secrets. Runs none of the code. Nothing sent yet. You approve the model, consent and cost 2 Understand Each file, then folders, then the whole system. 3 Write Each section for your audiences, real diagrams. 4 Check Looks up names. Links exact lines. Scores the guide. Guide ready Checked before you see it Paused at a model limit, or by you? Finished work is kept and never billed twice; resume on any model.

How it works · Security and privacy · How we check · Command line · Integrations · Connect an AI assistant · AI agents

Try it on a repository you know

Free to use with your own model key.

Start a guide