IceShore and still know what's inside
Your AI reads your Terraform or other IaC and your documents, then adds to your architecture diagram who uses the system, for what, and roughly how many. On the same diagram, people see the structure and AI sees the business context the code leaves out. IceShore helps reduce the number of systems no one understands well enough to maintain.
Send to my computer to start free
Vibe coding is outrunning your team #
AI agents write code faster than teams can follow. Systems get tangled, and three jobs fall behind: drawing the architecture, following each use case, and maintaining what’s running.
Build architecture diagrams from code, automatically #
Systems built with AI drift off course when nobody on the team understands them. IceShore draws them as the architecture diagrams developers already know.
Follow each use case on the diagram, step by step #
AI also writes piles of documents that nobody wants to read. IceShore shows one use case at a time on the diagram and numbers each step in order.
Ask your AI: “The expert who knew this API has left. Who’s affected if we shut it down?” #
Your AI reads the same architecture diagram your team does. To maintain a system well, it needs more than the code. It needs to know how the system connects to the business. IceShore gives it the use cases, so it can see that connection.
Maintaining systems now takes AI. #
Code that keeps growing
Copy-pasted lines went from 8.3% to 12.3%, duplicated blocks up 8x. Refactoring fell from 25% to under 10%.
Source: GitClear 2024
Dependencies and risk you can't see
Nobody can tell what a change will affect. Systems multiply until just finding them all is a job in itself.
Source: Spotify, on why it built Backstage
Not enough time
Developers spend 58% of their time understanding code, and 42% of the week dealing with technical debt.
Source: Microsoft Research / Stripe “What does this system actually do for the business?” No amount of code analysis will tell you, because it was never in the code. IceShore has your AI write it down as use cases, then gives them back to your AI whenever it needs them, so your decisions start from what the system is for.
Your AI gets better at system maintenance #
IceShore helps you understand your systems through architecture diagrams, and helps your AI understand your business through your use cases.
The next challenge in AI-assisted development is putting AI to work on system maintenance.
Startup CTO or tech lead
A small team shipping fast with AI, and running everything it ships.
Hand maintenance to the AI and the answers ignore business impact and dependencies.
Your AI answers from your use cases, and you keep shipping fast.
Enterprise dev or IT manager
Every department built its own tools with AI. You need the org-wide view.
No one can tell what a change will affect, so inventories and approvals stall.
Your AI answers impact questions org-wide, and credentials stay masked when you share.
Consulting architect
You maintain and extend legacy systems you inherited or someone else built.
No docs, no original authors. Ask the AI and it misses what the system is for.
Your AI drafts the use cases from your docs and code. You correct them, and it works from your corrections. Past projects carry over.
How your data is handled
- Your own AI does the processing. IceShore never sends your content to a generative AI.
- Credentials are masked before they are stored. Access keys, private keys, connection strings and passwords are detected and redacted, and each detection is recorded.
- Connections use OAuth 2.1 + PKCE. MCP is authorised with signed access tokens.
- You can read exactly what the AI wrote. The History tab opens each saved file version by version, down to the diff against the previous one.
- Hosted in the AWS Tokyo region. Stripe processes payments; we never hold card data.
What you can check for yourself
- The data format is public. An architecture is a text file in ADL (Architecture Description Language). The language spec and schema are published on GitHub under Apache-2.0, so what you export stays readable outside IceShore.Spec and schema→
- The setup steps are public. Claude Code, Claude Desktop, Codex, Cursor, Cline, GitHub Copilot and Windsurf, each with config you can paste as-is.Connect→
- You can see the real product without signing up. The screenshots on this page are the real sample workspace. You can open the same one.Open the sample→
Start free. Pay when your team joins. #
Free for one person. The more of your team uses it, the more your AI understands.
Getting even one new hire up to speed faster is worth tens of thousands of dollars (our estimate: the new hire's salary x 1.5 months, plus the salary of the existing members supporting them).
- For one person (no member invitations)
- Unlimited architectures, created from your own AI over MCP
- Change history and diffs
- Up to 3 guest viewers (view only)
[Get started free](https://app.iceshore.ai/?entry=signup)
[View samples, no signup→](https://app.iceshore.ai/workspace/sample/)
- Everything in Free
- Invite members and work together (up to your seat count)
- Guest commenters (3 per seat)
- Add or remove seats anytime, prorated. Starts at 1 seat.
- Download the member list as CSV
- Everything in Team
- 12 months of audit log history (30 days on Free and Team)
- Download the audit log as CSV
- Bulk member updates (CSV upload)
- Export access rights as CSV (with approval dates and last sign-in)
- Export architecture diagrams as Markdown (with SVG images)
Frequently asked questions #
Can't my AI just read the code and CLAUDE.md?
Often, yes, for questions that stay inside one repository. But the code can't tell your AI who uses the system, how many people that is, or which parts of the business depend on it. You can add that to a CLAUDE.md, but it stays inside the repository it lives in. With IceShore, your AI writes each system's use cases into its architecture, and can read them across the systems in your workspace.
Who writes the use cases?
Your AI does. When you ask, it reads the code, configuration files and documents on your machine, organises the use cases and, once you confirm, registers them with IceShore. Anything the code doesn't say, such as how many people use the system, comes from your documents or from you. The diagram is drawn automatically; fix anything wrong in the Correct tab. From then on, your AI works from your corrections.
Is my code sent to an AI?
IceShore never sends your content to a generative AI. Drafting and answering both happen on the AI you already use. IceShore acts as an MCP server, returning architecture information when your AI asks for it.
Where is my data stored?
In the AWS Tokyo region (ap-northeast-1). Stripe processes payments and we never hold card data. The details are in our privacy statement.
What if a config file contains credentials?
They are detected and masked before anything is stored. Access keys, JWTs, private keys, connection strings and password or token values are covered, on both the in-app save path and the MCP save path. Every detection is recorded.
Can I verify what the AI wrote?
Yes. In the History tab of your own architecture (editor or above), each saved file opens version by version, down to the diff against the previous one. Every version records whether it was written by the AI over MCP or edited by a person.
Which AI can I use it from?
Any AI client that speaks MCP. Authorisation uses OAuth 2.1 + PKCE with signed access tokens, so you are not locked to one AI vendor.
How far does the free plan go?
Free forever for a single person. Creating architectures, connecting over MCP and change history are all included. Paying starts when you invite members to work together. On the Free plan after a paid plan has ended, the members you already had can view and export the architectures, but no one can edit them.
Do people who only use IceShore through their AI need a seat?
Workspace members do: every member uses one seat, Readers included. People outside the workspace can be invited to individual architectures as guests, and guests don't use a seat. The guest limit counts across the whole workspace: up to 3 on Free, view only; 3 per seat on Team and Business, and they can comment. Guests can also read the architectures shared with them from their own AI over MCP.
What happens when I cancel or change seats?
Cancelling keeps your access until the end of the current billing period — nothing is cut off immediately. After that the workspace switches to the Free plan: you can still view and export your architectures, but no one can edit them. Members stay. You cannot invite anyone new, and any share link already issued stops working. Subscribing again restores editing. Adding or removing seats is prorated at that moment (increases are billed immediately; decreases are credited against your next invoice — they are not refunded).
Do you issue a Japanese qualified invoice?
Yes. We are a registered invoice issuer in Japan (registration number T7010001137750, Launchers Co., Ltd.). Invoices carrying the number are available in the app under "Invoices & payment method". The payment receipt emailed after checkout does not include it.
Toward a world where people and AI see the same ice field and move forward together. #
The age of AI-grown systems, multiplying without end. The countless systems we've created drift on like ice floes. IceShore wants to be the icy shore that gently moors them together, with AI.
[Get started free](https://app.iceshore.ai/?entry=signup)
[View samples](https://app.iceshore.ai/workspace/sample/)
Send to my computer to start free
[Sign up on this phone instead](https://app.iceshore.ai/?entry=signup)
[Read the full documentation](https://help.iceshore.ai/en/index.html)