BASECORD An Environment Where AI Can Shine
IDEAS COME ALIVE

BASECORD makes AI come alive likea fish in waterDevelopment quality and speed are on a completely different level!

An image of AI sharing memory and building together with Claude, Gemini, ChatGPT, and others in a workshop in the clouds
An image of AI working energetically, sharing tools and memory. The AIs shown are just examples.

BASECORD is where AI shows its true potential.
It has the tools to build, the memory to remember, and a place to run things.
So your "Can you build this?" starts to come to life.

AI × ENVIRONMENT

Give AI a workshop
where it can show its skills.
That's BASECORD.

An image of AI from Claude, ChatGPT, and Gemini building, testing, and refining a website in a well-equipped workshop

"I want to make something like this."
From that one sentence, files get created. They run.
Check the results, then refine again.

BASECORD is a development platform where AI works on a Linux server in the cloud. While you converse, it handles programs and images in the same workspace and builds things that actually run.

Like a fish swimming freely in water, we aim to give AI an environment where its strengths connect directly to the work. "Like a fish to water" is the phrase that captures the feel of using BASECORD.

01

What you build can run.

Not only can you write code and files, you can also run them on the server and check the results. Investigate errors and fix them, all through conversation.

02

Pick up where you left off.

Your goals, the approach you've decided on, and your past work. Continue from where you stopped, building on the same files that hold shared memory. This reduces the need to re-explain background every time.

03

Borrow another AI's perspective, too.

Claude, ChatGPT, Gemini, and other AIs work on the same platform with shared memory. When you get stuck, you can also consult a different AI.

CHAT WINDOW vs WORKSHOP

The same AI, but
"just talking" and "being able to work" are different.

The intelligence of the AI itself is the same, but the work it can do changes with what it has at hand. We've placed a regular chat screen side by side with BASECORD for comparison.

Left: an AI confined to a chat frame that can only talk. Right: AIs building apps and sites in a workshop, a contrast between the two.
TaskTypical chat screenBASECORD
Create filesCode is passed as text. Saving and placing it is up to the personThe AI creates files directly on the server and organizes them in place
Run and verifyWhether it works is unknown until a person tests itThe AI runs it itself, reads the errors, and fixes them
Check the appearanceCannot see the screenThe page is rendered as an image so the AI itself can check whether the layout is broken
PublishUsers' Servers Are Uploaded by PeoplePublish according to the steps you've decided, then check that it displays properly
Picking Up Where You Left OffWhen the conversation changes, you have to explain the background againPick up where you left off, remembering what you decided and the files you created
When you get stuckTend to repeat the same ideas with the same AIConsult other AIs such as Claude, ChatGPT, or Gemini from the same screen

"General chat screens" refer to using it by simply conversing in a browser. What you can do varies depending on the service and settings.

LINUX · TOOLS · SHARED MEMORY

The AI uses
BASECORD, a large
toolbox that controls Linux.

Just as a chef uses different kitchen tools, AI combines programs, databases, and image and video processing software. The person asking tells the AI "what they want to make," and the AI works out the tools and steps needed for the job.

An image of Claude, Gemini, and ChatGPT finishing apps and sites like dishes in a fully equipped kitchen
Memory, which serves as the shared recipe, and the tools available to use. Once the environment is set up, the work of turning ideas into reality moves forward.

BASECORD's AI combines and uses these tools on its own to produce the results it's been given.

"Modules" are the parts that add functionality. Linux offers a vast range of options, not only for such parts but also for languages, server software, and creation tools.

Deliver

Apache

A web server that delivers your created web pages to the browser. It forms the foundation for publishing your site.

Apache official site →Dedicated Server Plan
Process

PHP

Receives form input and builds screens based on conditions. You can write server-side processing for reservation intake and business applications.

PHP official site →Dedicated Server Plan
Store

MySQL

A database that organizes and saves products, reservations, inventory, and other data. It handles searching and updating the information you need.

MySQL official →Dedicated Server Plan
Assemble

Python · Node.js

Aggregate data, organize files, and build app processing. These are the tools for running programs that fit your purpose.

Refine materials

FFmpeg · ImageMagick

Convert video and audio, and resize or change the format of images. Processing the materials needed for production can also be done in the same workspace as your code.

Verify & revert

Browser testing · Git

Open screens to check the display and operation, and keep a record of changes. Supports post-creation verification and review when problems occur.

Deliver quickly

Nginx · Caddy

A web server that handles heavy traffic lightly. Used in place of or alongside Apache, chosen to match display speed and architecture.

Dedicated Server Plan
Protect

Let's Encrypt · SSH

Certificates that encrypt your site with https, and a secure way to log into the server. This is the protection that underpins publishing.

Dedicated Server Plan
Deliver to the world

Cloudflare

Pages are delivered from a worldwide distribution network, and stale cache clearing is also done via API. I used these every time, even during this one night.

Store in another form

PostgreSQL · SQLite

From a full-scale database to one that runs as a single file. We choose what fits your purpose.

PostgreSQL is a dedicated server plan feature
Learn it in an instant

Redis

Keep frequently used data in memory to speed up page display. It supports busy sites.

Dedicated Server Plan
Search

Elasticsearch

Build a search system that quickly finds documents and products from a large volume using words.

Dedicated Server Plan
Create images

Pillow · WebP conversion

Drawing images, adding text, cropping, and converting to lighter formats. This is the groundwork for making pages faster.

Recognize images

OpenCV · Tesseract

Find shapes and faces in photos, and read text inside images. It can also be used to digitize paper documents.

Listen

Whisper

Transcribes the audio of meetings and videos into text. It's a starting point for creating minutes and subtitles.

Speak

VOICEVOX

Reads text aloud with a voice. It can be used for video narration and apps with read-aloud features.

Dedicated Server Plan
Make documents

Pandoc · LibreOffice

Converts text into the formats you need, such as Word, PDF, and slides. You can automate the creation of reports and materials.

LibreOffice is a dedicated server plan feature
Handle PDFs

Ghostscript · Poppler

Reduce PDF size, split or merge pages, and extract text from them. Let AI handle your document work.

Show it in 3D

Three.js · WebGL

3D screens that run inside the browser. The renovation floor plans and the 3D Othello were built with this.

Build in 3D

Blender

Software for making 3D models and videos. AI can drive it with commands and render the results without you opening the screen.

Dedicated Server Plan
Test on screen

Playwright · Chromium

Opens pages in a browser with no screen, clicks, and takes screenshots to verify. This is the tool that serves as AI's "eyes."

Dedicated Server Plan
Test it

pytest · Jest

Automatically checks, again and again, that the programs you build work correctly. Prevents breakage after fixes.

Analyze

pandas · Matplotlib

Totals up sales and inventory tables and turns them into charts. Quickly creates material for making decisions from the numbers.

Teach it

scikit-learn · PyTorch

A mechanism that learns trends from data. Useful for demand forecasting and automating classification.

Turn it into an app and distribute it

NSIS · PyInstaller · PyQt

Builds Windows app screens and packages them into installers for distribution. SlideShow Pro was built with this.

Publish and keep it

GitHub API

Upload the files you built to GitHub and create a release for download. It also serves as the place where downloads are hosted.

Pack it in a box

Docker · Podman

Pack the app together with the environment it runs in, so it runs the same way on any other server.

Dedicated Server Plan
Run it automatically

cron · systemd

Nightly aggregation, regular cleanup, and automatic restarts when something stops. Work keeps going even when no one is watching.

Dedicated Server Plan
Keep watch

fail2ban · ClamAV

Shut out unauthorized login attempts and scan files for viruses. Strengthen your server's defenses.

Dedicated Server Plan
Connect to the outside

Email · payments · each company's API

From sending email and payments to Google services and other AIs. Combine with outside systems to expand what you can do.

A user puzzled and frightened by the many unfamiliar software names floating overhead
Eh!? What's that???

Honestly, I have no idea what any of that means...

But don't worry!

You don't need to understand any of the complicated technical things. The AI will pick the tools it needs on its own and combine them to answer your requests, and it will know how to use them all!

AIs that combine tool cards like a card game to create apps, sites, and games
Tools are the AI's hand of cards. It chooses the best combination for each request and shapes it into something real.

The 30 cards above are just the entrance.

Linux tools come in a completely different scale. The two on the left are counts we actually tallied on the server that runs this page. The two on the right are the sizes of public component libraries anyone can use.

About 36,000Parts you can install right away from the OS's official repositories
1,296Commands you can use on this server right now
Hundreds of thousandsRegistered in the Python package index (PyPI)
MillionsRegistered in the JavaScript package index (npm)

When needed, the AI looks things up on the spot, installs them, and checks how they work. Below is just a small part of that shelf (83 items). Gold (★) marks what was actually used in this overnight work.

Publish & Deliver

ApacheNginxCaddyCloudflare APILet's EncryptSSHrsyncPostfix (email)

Languages

PythonNode.jsPHPBashGoRustJavaC / C++ (GCC)RubyPerl

Store Data

MySQLMariaDBPostgreSQLSQLiteRedisMongoDBElasticsearch

Create & Fix Images

ImageMagickWebP conversion (cwebp)PillowOpenCVInkscapelibrsvgpngquantoptipng

Video & Audio

FFmpegSoXLAMEyt-dlpWhisper (transcription)VOICEVOX (text-to-speech)

Documents & PDF

PandocGhostscriptPoppler (PDF to text)qpdfLibreOfficeTesseract (text recognition)Mermaid (diagrams)

3D & Games

Three.jscannon-esWebGLBlenderPeerJS

Verify on Screen

Chromium (headless)PlaywrightPuppeteerpytestJestLighthouse

Data Analysis & AI

pandasNumPyMatplotlibscikit-learnPyTorchJupyter

Turn it into an app and distribute it

NSIS (installer)PyInstallerPyQtElectronGitHub APIGitDocker / Podman

Monitor & Automate

cronsystemdjournalctlfail2banClamAV (virus scanning)curljq

Connect with Other Services

Gmail / SMTPGoogle Drive APIStripe (payments)SlackLINEAI APIs from each provider

Say, "Build a reservation management app for us."

  1. Build the screens: Prepare a form where users enter the reservation date and number of guests.
  2. Connect the processing: Validate the input with PHP and save the reservation to MySQL.
  3. Deliver to the browser: Serve the screens from a web server such as Apache.
  4. Run and fix: Test registering, changing, and displaying reservations, and fix any problems found.

This is an example configuration for a dedicated plan. Depending on your goals, different tools may be chosen. We build each request as a system where multiple tools work together.

The more tools there are, the more ways to build.

Linux has a rich set of long-established software. AI looks up how to use it, prepares the necessary settings and code, and checks the results. BASECORD is the workspace where that work can be carried out through conversation.

Not every tool listed here is installed by default. What you can use depends on your plan, installation status, permissions, and server performance. Shared plans cannot run PHP or database servers, and use a storage-only API for data. Dedicated plans let you choose configurations that use PHP, MySQL, and other tools.

BEHIND ONE SENTENCE

What AI does
behind a single request.

This is a real exchange. The creator sent only one screenshot and this one line.

An image of AI agents relaying research, image creation, packaging, shooting verification, testing, and uploading, all behind a single line sent from a smartphone
"The right side of the installer screen still looks jagged."
AI's reply (summary):
The cause was that small images were stretched to fill the frame. I prepared 102 images at different sizes, measured the actual size of the frame, and changed it so that the exact-fit image is placed at its native size. I bumped the version by one and published it.
  1. Find the cause Read how the installer is built and pinpoint where images get stretched
  2. Remake the assets Draw 102 images at different sizes with Python and match the typeface to the logo
  3. Check with your own eyes Render before-and-after images side by side and confirm the jagged edges are gone
  4. Rebuild Reassemble the Windows installer and bump the version number by one
  5. Distribute Upload the new version to the distribution site and update the download button's destination
  6. Cross-check Re-fetch the distributed files and verify they match the originals using checksums
  7. Remove stale copies Purge outdated pages still cached on the delivery network, and make old versions impossible to fetch
  8. Report Separate what was fixed, what was verified, and what has not yet been verified

The creator's only role was to look over the work and point out anything that seemed off. The remaining eight steps were carried out by AI combining Linux tools.

The tools the AI actually used that night

PythonImage processing (Pillow)Windows installer creation (NSIS)Screen capture with a headless browserFFmpegGitGitHub APICloudflare APIMySQL (text in 12 languages)SHA-256 verificationA separate AI for translation

"In its element" means having access to these tools and being able to check the results yourself.

CONNECT · CREATE · AUTOMATE

Linux tools, connected to services worldwide.
With APIs, the scope of work reaches beyond the server.

It isn't limited to the tools on the server. Through APIs, the "contact points between services," external functions and data can also be part of the work. The AI writes the connecting programs, receives the information it needs, and passes it on to the next task.

An image of multiple AIs working together in a bright cloud workshop, connecting to services for images, video, calendars, email, maps, and business data
An image of the AI workspace connecting with external services. The services shown in the image are examples of use cases, not a built-in list of integrations.

For example, it can connect to work like this.

IMAGE

Create images and place them on a page

Request images from an image generation service and receive the finished files. Adjust the size and format, then incorporate them as material for product introductions or websites.

VIDEO & VOICE

Create videos and narration

Connect to video generation and speech synthesis services to prepare materials. Combine them with editing tools on Linux to finish a presentation video with subtitles and audio.

MAIL & CHAT

Notify and compile communications

Work with email and chat services to summarize inquiries and draft replies. You can set up a flow that sends after approval, or notifications under conditions you define.

CALENDAR

Link to schedules and reservations

Check free time from an authorized calendar, propose reservation options, and set up a system that registers confirmed details on the calendar.

MAP & WEATHER

Use location and current information

Retrieve map, route, and weather data to build screens useful for store directions, travel plans, and outdoor event preparation.

BUSINESS DATA

Combine sales and inventory

If the connected service supports it, import data from online sales or customer management systems for totals and reports, reducing manual data entry.

DOCUMENTS

Turn documents and spreadsheets into work

Read files and spreadsheet data stored in the cloud to create reports and materials. You can also set up a flow that returns deliverables to an approved storage location.

PUBLISH & DELIVER

Publish and distribute finished products

Register files with a distribution service, then update delivery settings and cache. Not just building, but also delivering, can be handled in one place.

YOUR SYSTEM

Connect to your existing internal systems

If your in-house system has an API, its features can be candidates for connection. Link internal information with the new screens you build to fit your daily work.

"Summarize this week's sales and create a report for review"

  1. Receive via API: Retrieve sales data for the target period from an authorized sales service.
  2. Process with Linux tools: Aggregate the data and create charts and a report.
  3. Review the content: Check the figures and text, and turn them into a draft that people can verify.
  4. Approve and deliver: Store it in the designated location and, if needed, share it by email or chat.

An API is the gateway that connects to outside services. Linux tools are the workbench that processes what you receive. On BASECORD, you can build the procedures and programs that use both by talking with AI.

External services are also connected in the making of this page.

In addition to generating the images shown, we used the GitHub API to register published files for the SlideShow Pro distribution, and the Cloudflare API to purge old cache when updating the web page. This is a real-world example of linking external functions with server-side work.

The above are examples of systems that connect and are developed. Not everything works out of the box; settings depend on the other side's API availability, contracts, authentication, permissions, and fees or usage limits. Separate from your AI chat subscription, external API usage fees may apply.

Keep secrets such as API keys out of public pages, and limit the data sent externally and the operating permissions to what is necessary. Build continuous automated processes with defined execution conditions. Shared plans cannot publish or keep server-side programs running, so if an app you build needs an endpoint that uses a secret key, the design calls for a dedicated plan or another secure execution environment.

BUILD · CHECK · PROTECT

Strengthen security with AI.

Turn your building power into protective power too. You can ask multiple AIs, including Claude, ChatGPT, and Gemini, to review code and settings, fix problems, and verify the fixes afterward. They share the same memory and files while bringing different perspectives, building up defenses step by step.

An image of many AI robots working together inside a building to install a massive defensive panel in preparation for outside attacks
Inspect, seal the holes, and verify again. This is an image of collaborative work by multiple AIs, likened to building a protective wall.
“Before publishing this app,
check for exposed personal information or keys, and any potential security vulnerabilities.”
01

Look for oversights before publishing.

Check names, addresses, school names, API keys, and links to admin pages. Confirm that no information that doesn't need to be public remains in files or on screens.

02

Review the code and settings.

Check whether any input or imported file could lead to a path where malicious programs run. Also examine access permissions and the components in use, and make the necessary fixes.

03

Even after fixing, verify.

Confirm the fix with safe tests that reproduce the problem, and check that the original features still work. Ask another AI to re-check, then organize what was confirmed and the remaining issues.

Going from review to fix in actual production work too.

In the Renovate Simulator in the production list, we reviewed the display processing for imported furniture data and found a problem where malicious programs could run. We applied a fix that displays furniture names as plain text and added checks on imported content, and confirmed the fix with test data. In the learning app, we checked school names, personal names, and similar details before publishing.

InvestigateCheck the publishing scope and the problematic processing.
FixKeep a backup of the pre-change state and correct the necessary parts.
TestConfirm the problem is resolved and the original features work.
ReviewRe-check from a different perspective and record remaining issues.

AI-based review is support for improving safety. Even with multiple AIs, oversights can happen, and there is no guarantee that every attack will be prevented. Regular updates and backups are also important. The scope of review and fixes you can perform depends on your plan, permissions, and connection settings.

YOUR CLOUD WORKSPACE

Take your time on a PC.
On your phone, whenever an idea strikes.

Your workspace is in the cloud. Open it from a browser and tell it what you want to build or what you want to fix.

Continue ideas from your home PC, or make fixes you think of while out. Work with the same environment's files and conversations.

The AIs you can use depend on your own subscriptions and connection settings. You don't need to have a specific number of AIs set up.

An image of using BASECORD on a desktop PC and a portrait-oriented smartphone
A device image we created earlier. Fine details on screen may differ from the actual display.
Actual operation screen
An actual screen of consulting on system development in BASECORD, with a conversation area, an input field, and an AI selection area lined up

Send a request from the bottom of the screen and check the AI's replies and work. Click an image to enlarge it. The display reflects the state at the time of capture.

FROM WORDS TO WORKING THINGS

Start with
"I want it to be like this" rather than technical terms.

For example, your first request can start with words like these.

“Create a website for a fictional hot spring inn.
I want an appealing site featuring an open-air bath by the river.”
STEP 01TellDescribe your purpose or image in your everyday words.
STEP 02Have it builtThe AI creates the files and screens you need.
STEP 03Look and tryCheck the pages and apps it made yourself.
STEP 04Refine and grow"Make it brighter", "make this button bigger".

You don't need to prepare perfect instructions from the start. Look at what takes shape and move it closer to your taste and how you use it. That back-and-forth is what development is.

The request above is a shortened version of the prompt from a production example. It doesn't always finish in a single request; it needs checking and revision depending on the content.

A REAL LOG

What you're told
becomes the next version.

"SlideShow Pro", a slideshow creation app for Windows. Some excerpts pulled from a single night's exchanges.

  1. BUILD 204Stop installing through PowerShell. Have it install through a dedicated screen.Reworked into a wizard-style installer that proceeds with [Next].
  2. BUILD 205Make the range bar and the playback position bar the same length and in the same position.Before building, drew the understood content as a diagram to confirm. Redrew it, then implemented.
  3. BUILD 207Lots of deletion text scrolls by, and it makes me a little uneasy.Replaced with a one-line description and a progress bar that moves. Deletion itself is also made faster in bulk.
  4. BUILD 208Put each BGM track in its own frame so it fits even when narrower.Based on a layout proposal drawn by another AI, added feedback and finished.
  5. BUILD 211Moving the volume bar during preview doesn't change the volume.Identified the cause and fixed it. Moved the preview and confirmed that 30% produces 30% volume.

Read the production story of SlideShow Pro →

SHARED MEMORY

Work with AI that “remembers,”
instead of starting from scratch every time.

BASECORD's AI reads its rules and memory before it starts any work. No matter which AI you ask, it begins from the same foundation.

LAYER 1

Rules for the Whole Project

Policies shared across every workspace. Say it once, and it applies to everyone.

  • Make the design bright and easy to read
  • Write URLs in full, in a form that can be opened
  • Convert large images to lightweight formats
LAYER 2

Instructions for This Job

Agreements for each workspace. These are the project's handoff notes.

  • When you fix and redistribute, always increase the version number by one
  • Run it through a checking tool before publishing
  • Translate only when asked to "translate"
LAYER 3

Index of Past Conversations

A bookshelf that summarizes and keeps earlier work. It can answer "Wait, where did we leave that?"

  • Where Previously Made Games Are Kept
  • Past decisions and the reasons behind them
  • How the work came about

How memory scope and sharing work depends on your plan and settings. The examples are ones actually used in creating this work.

DELEGATE WITH CONFIDENCE

Let it work on its own,
and you can undo it. You can verify it.

Because the AI does the work itself, use it together with a system that lets you roll back changes.

Keep a backup before changing

Before working, a copy of each file is saved before it gets rewritten. If you don't like the result, you can restore it.

Changes are logged in history

Changes to the workspace are recorded automatically, so you can later trace when and what changed.

Review Before Publishing

Before publishing, we check the site with tools for names, addresses, keys, and any openings that could let malicious programs in from outside.

Separate "What We've Verified" from "What Remains"

In reports, we clearly separate what has actually been verified from what has not yet been checked. We never call something "complete" if it hasn't been verified.

The systems introduced here are examples of how we operated while building these works. Available features vary depending on plan and settings.

MADE WITH BASECORD

Something born from words,
see it for yourself.

These works were created by makers who are not engineers, through conversations with AI.
Not only the finished results, but also the inspiration and the words used to request them are shared here.

Website Examples Built with AI

Turning an imaginary shop into a site.

A riverside hot spring inn, luxury cars, a private jet. A portfolio of websites that turn imaginary worlds into images and pages.

View the Website Portfolio →
3D floor plan screen of Renovate Simulator

Imagining a renovation, in 3D.

Thinking through interior layouts and furniture placement in advance. A 3D app made in a single day as a side project. It is a prototype still under development for practical use.

View the Production Story →
An image of using a smartphone study quiz on the commute to school

Turning a family's troubles into an app.

A study app born from a everyday wish: to memorize exam material efficiently on the train to school. It teaches through pronunciation, diagrams, and 3D.

Study App Production Story →

The works currently published are test pieces to try out what is possible. The shops and inns on the website are fictional.

A LITTLE MORE ABOUT BASECORD

For First-Time Visitors.

Can I use it even if I can't program?

Just make your request in natural language, and AI will build the code and screens for you. Stating your goal, testing the result, and explaining anything that bothers you are important roles for the person using it.

Does the result change just because the AI is different?

Besides the type of AI, the tools available, your instructions, the work environment, and the accumulated checks all affect the outcome. BASECORD brings that work environment together in one place. It does not guarantee higher quality than other services for every request.

Are there differences in what you can build or which features you can use?

Yes. Shared plans do not support server-side programs such as PHP, but you can save your app's data to the server with the storage-only API. Dedicated plans (one server per company) also allow development using PHP, databases, and more. The AI and image generation tools you can use also vary depending on your contract and settings.

Is it okay to publish something right after the AI creates it?

Before publishing, you need to check its operation, personal information, rights, and security. Use AI-based review as well, and verify and refine the result yourself.

Will the AI create too much on its own or break things?

As a basic rule, it only fixes the scope you asked for and keeps a backup before making changes. We confirm before any irreversible operation or work beyond the scope you requested. Even so, please always verify the results with your own eyes.

GIVE YOUR IDEAS A PLACE TO GROW

Where AI shines.
A place where your ideas start moving.

"I wish something like this existed."
Why not shape that idea together with BASECORD?