IngestMD: The local-first AI context builder.

Local-first AI context

Give AI the right context, without leaking your code

Describe a task and IngestMD packs the exact files the model needs, secrets redacted and nothing uploaded. Discover areas, review diffs, and ingest documents, all to a token budget, all on your machine.

IngestMD building task-scoped context for a job

For consultants, legal, and finance

Analyze a confidential folder, privately

Bundle an entire client folder and reason over it with a local model. Nothing is uploaded, nothing leaves your machine.

IngestMD bundling a folder into Markdown

Stop dumping whole repos

Hand the AI exactly the files the job needs

Describe a task and IngestMD ranks, expands, and packs the right context to a token budget, with a quality score.

IngestMD Task context

For large codebases

Load just the subsystem you are fixing

Areas splits the repo into its real parts. Serve one over MCP instead of drowning the model in everything.

IngestMD Areas discovery

Beyond code

Turn a Confluence space into a NotebookLM briefing

Pull pages into clean Markdown, then let NotebookLM make an audio overview or a deck.

IngestMD Documents and Confluence

Your own model

Ask your codebase, answered on your machine

A token-budgeted context goes to your local LLM, and an answer comes back, with no cloud round-trip.

IngestMD Ask answering a question
Download for macOS, Windows, Linux View on GitHub

Free · Open download · 100% local · No telemetry

234,610files scanned
5,851bundles
~23sto finish
100%offline

The annoying part

NotebookLM gives you a limited number of sources. Uploading a repo file-by-file wastes all of them.

Feeding a project to an AI tool means flattening the tree, skipping binaries and dependencies, watching for leaked keys, and staying under a token budget. IngestMD does all of that in one click, so a whole folder becomes a handful of clean files you can drop straight in.

What it does

Everything you'd otherwise do by hand

One file per folder

Every source file embedded verbatim in fenced code blocks. Large subtrees split into parts. No more dragging files in one by one.

Secrets redacted

API keys, tokens, and private keys are scrubbed before anything is written, so you never paste a credential into a chat.

Skips the junk

Binaries, node_modules, build output, and anything in your .gitignore stay out. Add your own exclude patterns too.

Token counts

Every bundle shows an approximate token count, so you know the context cost before you load it into a model.

Local by default

Discovery, bundling, and redaction run on your machine, offline, with no account or telemetry. The only network egress is opt-in: your own local LLM, the optional Pro cloud models, and Confluence retrieval.

MCP server included

Eight tools over MCP for Claude Code, Cursor, VS Code, or Codex. List the areas in a repo, load just one, name them with your local model, or pull task and diff context. Ask in plain language and the AI picks the tool.

Ask your codebase

Point IngestMD at your own local LLM server (Ollama or LM Studio) and ask questions about a repo. It builds a token-budgeted context and answers on your machine, no cloud, no API key.

Task-scoped context ✦

Describe a task and IngestMD assembles exactly the files you need: semantically ranked, dependency-expanded, packed to a token budget, with a signatures-only skeleton fallback. Plus a quality score and a "why each file was included" manifest.

Review your changes ✦

Turn your current git changes into a review-ready bundle: changed files, their tests, and their dependents, behind an AI review prompt. Diff against your working tree or any base branch.

Discover areas ✦

IngestMD clusters a repo into its real subsystems by imports and shared terms, then writes one titled, budget-packed bundle per area. Deterministic and collision-free across large, small, and Salesforce projects. Focused context instead of a flat dump.

Curate a Collection

Build a cart of bundles, areas, answers, and task or review contexts, then export a tidy set of Markdown with an index. Save collections by name to reuse, or serve them to AI clients over MCP.

Ingest documents & Confluence ✦

Turn PDFs, Word docs, and spreadsheets into clean Markdown. XLSX sheets become tables, DOCX keeps its headings, secrets are redacted like code. Pull Confluence Cloud pages, spaces, or CQL queries straight into Markdown too.

Find exactly the files ✦

Search your repo by name, extension, glob, language, size, or content (literal, regex, or semantic), then add the matches straight to a collection. Stop dumping whole folders, and hand the AI precisely the files that matter.

Name and explain your areas ✦

Have your local or a low-cost cloud model give each area a clear title and a plain description, then generate a deeper brief (purpose, key files, data flow, entry points) when you want to understand one fully. Saved, searchable, and shared with the MCP server.

Scope a cloud AI to one area

Copy a ready-made prompt that locks a cloud AI to a single area with hard boundaries: answer only from these files, and ask before crossing into another. Self-contained for any chat, or MCP-aware so it can fetch more after you approve.

Cloud models, your key ✦

Prefer a hosted model for enhancement? Pick a low-cost one (OpenAI, Mistral, xAI, Gemini, DeepSeek, or any OpenAI-compatible endpoint). Your key stays in the OS keychain, and a cost estimate shows before you run.

Save projects

Save a source folder and all its settings as a named project, then load it later to rescan. Start fresh in one click, and choose local or cloud per project.

See the full feature breakdown, by capability »

Understands your code

It maps how a codebase fits together, not just what is in it

Tested on 21 real open-source projects plus Salesforce orgs, fully offline and deterministic. It finds the natural areas, follows the imports, and connects the components, with the same result on every run.

150,542files mapped
21real codebases
1,277areas found
10,398connections mapped
10languages
100%offline

Reads Python, JavaScript, TypeScript, Rust, Go, C, C++, C#, Dart and Flutter, and Salesforce Apex with SOQL.

On Salesforce it goes further: it groups LWC, Apex, and objects into the features they form, following @salesforce/apex calls and SOQL into your data model.

See the full benchmarks, every repository »

A real desktop app

Built with Tauri + Rust. Fast, native, tiny.

Every screen, shot on a real repository.

Clean Markdown bundles, ready for NotebookLM, ChatGPT, or Claude.
Clean Markdown bundles, ready for NotebookLM, ChatGPT, or Claude.
Task context (Pro): describe a job, get the files that matter, packed to a token budget.
Task context (Pro): describe a job, get the files that matter, packed to a token budget.
Diff review (Pro): your git changes, their tests, and dependents in one bundle.
Diff review (Pro): your git changes, their tests, and dependents in one bundle.
Area discovery (Pro): the codebase split into coherent subsystems.
Area discovery (Pro): the codebase split into coherent subsystems.
Ask your codebase (Pro): answers from a local LLM you run, nothing leaves your machine.
Ask your codebase (Pro): answers from a local LLM you run, nothing leaves your machine.
Search (Pro): find files by name, type, language, or content.
Search (Pro): find files by name, type, language, or content.
Documents (Pro): PDF, Word, and Excel to Markdown, plus Confluence.
Documents (Pro): PDF, Word, and Excel to Markdown, plus Confluence.
Collection: curate exactly what you hand an AI, then export it.
Collection: curate exactly what you hand an AI, then export it.

How it works

Four steps, about ten seconds

  1. 01

    Pick a folder

    Choose a source project, and optionally a separate output folder.

  2. 02

    Tune (optional)

    Toggle redaction and .gitignore, set grouping, add exclude globs.

  3. 03

    Bundle

    One click. IngestMD scans once and writes clean Markdown.

  4. 04

    Drop into your AI tool

    Load the .md files into NotebookLM, ChatGPT, or Claude.

Use cases

More than code. Any pile of files, kept private.

For engineering teams

Onboard a new developer in days, not weeks

Hand a new hire a guided tour of the codebase their AI assistant actually understands.

For any team on Confluence

Turn a Confluence space into a NotebookLM briefing

Pull a whole space into clean Markdown, then let NotebookLM turn it into an audio overview or a deck.

For consultants, legal, and finance

Reason over a sensitive client folder, privately

Bundle an entire confidential folder and analyze it with a local model. Nothing is uploaded.

See all use cases »

For AI clients

An MCP server, built in

IngestMD ships an MCP server over stdio with eight tools. Free: bundle, discover_areas, list_areas, area_context, and enhance_areas (area naming runs in the desktop app; the MCP serves the result). Pro adds task_context, diff_context, and saved collection. Point a client like Claude Code at it and load a repo, or just one area, as live context, with the same scanning and redaction as the app. See the full MCP guide and prompts. 

{
  "mcpServers": {
    "ingestmd": {
      "command": "/path/to/ingestmd-mcp"
    }
  }
}

Why IngestMD

More than a repo dump

Tools like Repomix and GitIngest are great when you want to flatten a whole repo into one file fast. IngestMD is built for the next problem: handing AI the right context, safely, without dumping everything and burning your token budget.

A flat repo dump

  • The whole tree in one file.
  • You hand-tune include and exclude rules.
  • The model still gets too much, and the wrong parts.
  • Secrets and binaries are your problem to catch.

IngestMD

  • Task-scoped and area-scoped context, packed to a token budget.
  • Secret redaction and junk skipping on by default.
  • Area discovery, collections, and a context quality score.
  • Local-first with no telemetry, served to AI clients over MCP.
  • Salesforce-aware, with Pro and Team governance for client work.

See the full comparison, by capability »

Pricing

Start free. Upgrade when context gets serious.

The free bundler is fully featured and stays that way: safety is never paywalled. Pro unlocks intelligent, task-scoped context. Every download starts with a 3-day full Pro trial, no account and no card. Local-first: unlock later with a license key.

Free

$0 forever

The deterministic bundler, trustworthy on its own.

  • Bundle any repo to clean Markdown
  • Secret redaction + binary safety (never paywalled)
  • .gitignore respect · token counts
  • MCP bundle tool
  • Curate & export a Collection
Download
Launching soon

Pro

$9 /mo · or $99 lifetime

Intelligent task context: the right files for the job.

  • Everything in Free
  • Task-scoped context: rank → budget → skeleton
  • Diff review bundles from your git changes
  • Area discovery + Ask your codebase
  • Context quality score + recipes
  • Policy & safety profiles, saved collections
  • Salesforce mode
Checkout opening soon
Launching soon

Team

$19 /user/mo

Governed context: consistent, reproducible, audited.

  • Everything in Pro
  • Shared profiles, recipes & policies
  • Approved, team-safe context packs
  • CI generation of context bundles
  • Audit log of what was sent to AI
  • Admin / MCP / repository policy controls
Checkout opening soon

Prices shown are introductory and may change. Every tier runs entirely on your machine.

Get it

Download IngestMD

macOS Apple Silicon · Intel Windows .exe · .msi · portable .zip Linux AppImage · .deb · .rpm

Free and open to download. The macOS build is signed and notarized by Apple. Windows builds are not yet signed, so on first launch click "More info" then "Run anyway". Prefer no install? The Windows portable .zip runs straight from a folder, on any machine with the Edge WebView2 runtime (already on Windows 11 and current Windows 10). All releases »