on a more serious note, everyone should have a little utility belt of lists, hand curated and intimately known. they are an extension of me at this point.
I see the other comment is marked dead now, but seriously why? I'm having a hard time imagining how a schema is viewed as actual intellectual property or a security risk.
Intellectual property: it represents what kinds of data you maintain, their naming, and their relation. That’s valuable on its own and also points to what your software does.
Security: well, see above—knowing what data is stored determines your value as a target.
Now is it the _best_ way to target an adversary? Probably not.
I've used DBML and like it as well. There's a small ecosystem of tools that support it, and I think the additional structure with curly brackets makes for fast "eye parsing".
dbdiagram is definitely an useful utility. Some of the differentiators would be:
1. Simpler Syntax
2. Fully Local processing (This tool works offline, once you load the original web page)
3. Light weight & fast (the entire project is 20kb in size)
It does look VERY similar, but I think they tried to make the language a bit more terse. As a said is another comment, I think the slightly additional structure is actually better.
From what I can tell, a lot of these tools seem OK for capturing relatively simple database designs, but at least "look" fairly inadequate for more robust needs. This makes sense if you're trying to use a GUI for creating a database: there's going to be limits to how much you can accomplish easily with that approach so being robust for smaller databases works out.
Most of the time I want an ERD tool it's to capture a database that I've already built for documentation purposes, or to document a database that someone else has built and that I need to come to terms with.
For this I use SchemaSpy (https://schemaspy.org/). I let it reverse engineer the database, build the diagrams, etc. I like the way it organizes what it captures and that it will do things like capture in-database object comments and put them in sensible places; for example, this lets me use the PostgreSQL COMMENT ON * commands to add contextually relevant documentation in the DDL and this tool will do the right thing when I generate database documentation.
I used SchemaSpy to great effect for many years, till someone decided it was abandoned and took it over. When I tried again it had become shockingly bloated
presumably in the name of modernity. I just wrote a one pager for the SQL dialect I was in and moved on.
I get that it is java, and bloating is considered normal but as the original author proved, it does not have to be that way all the time.
This could well be the case, but to be honest I'm not too concerned by that. So in short, it may not be a well engineered product, but it seems to do its job well enough which is all I really care about as a user.
My usage pattern is to run it on a workstation (usually my personal workstation or laptop) and take the result and put it somewhere useful. It's not a perpetually running process or something I'd run on, say, a production machine where any bloat could produce meaningfully negative results. Those results are also not something I'm too worried about for things like bloated web pages or so... at least insofar as the audience for that result is fairly limited and the user experience isn't something I see as hurting usability. Any extra time it takes doing its job because of any inefficiency doesn't really impact me either; once I'm at a point I can set it running on a database, I haven't found it to be terribly long running, I don't run it often, and usually I can just do other stuff while it's running (a few minutes or so).
I haven't found anything open source that produces as good a result and its commercial competitors aren't any better in any sufficiently meaningful way... unless you want to start talking about database documentation being integrated into larger, consolidated requirements/documentation tools.
If anything I'm pleased that someone did pick up the maintenance and that new database features and the like have some hope of being supported... and without maintenance the result -would- eventually come to suffer as things in the database world moved on.
The issue I have with a lot of these tools is they work fine when depicting relationships between tables in the same schema (talking mainly about PostgreSQL databases), but few support showing relationships between tables across different schemas.
Also, when the number of tables grows large, few have layouts arranged in an optimal way. I use D2 (https://d2lang.com/) to create ERDs. However, of the free layout engines available in D2, Dagre (https://github.com/dagrejs/dagre) and ELK (https://github.com/eclipse/elk) both don't have optimal placement of layouts for a sufficiently complicated database.
I've used Mermaid, partly because it integrates well in both GitHub and Notion. It's fine, quite featureful really, but the lack of any ability to control layout properly makes complex schemas quite a mess when rendered. Not sure how others of these declarative, text-based engines deal with that.
Oh!
I thought this was something about describing a data model in plain text and having an LLM generate the entity-relationship diagram tool.
I’ve done this with GPT4, asking it to output Mermaid relationship diagrams. And it’s usually impressively good (never went beyond ~10 entities). I also had it update and fix the result using natural language feedback.
I sometimes had to touch up a few things "by hand", but I really wouldn’t bother doing it from scratch myself anymore.
Even the failures are interesting, as they often result from an unclear description, or issues in my own understanding of the domain or model.
ditto, GPT-4 is very impressive with all types of mermaid [0] diagrams! even helps w/ styling and formatting tweaks. It's helped me so much with better coverage and maintainability on diagrams for software & systems projects.
I love that this is trending on HN. LLMs be damned. We still have lots of “simple” problems to be solved, like how to effectively write and share an ERD.
A LLM would probably make this better. Instead of asking the user to learn a DSL why not just translate SQL to whatever structure the render functions uses?
Great ! I am also excited to share ERDTOOL built in this space - https://erdtool.com ...
- Multiplayer entity-relationship-diagramming tool
- think Figma of ERDs
- Supports Postgresql and MySQL (mysql part is new homepage hasn't been updated to reflect that)
Building it solo.
I hope this will also interest the tech community
My own tool adopts a mixed approach to the layout problem : diagram-as-text, but the line breaks and the order of the boxes (entities and relations) are used as position hints. The resulting layout is constrained to an invisible grid corresponding to a left-to-right and top-to-down reading of the source.
Moreover, since there are relatively few possible grid positions, automatically rearranging the layout becomes an assignment problem, which can be solved by a simple b&b algorithm.
Obviously, this wouldn't scale, but this is just a teaching tool that I use to present small ERDs to my students.
You may test it at https://mocodo.net. Click the turning arrows symbol to draw the diagram and the shuffle symbol to rearrange it randomly.
I am using ChatGPT to generate PlantUML Entity Diagrams. I have a separate chat just for this purpose. It can generate the PlantUML and Spring JPA Java code for all my entities. If I need a change , I ask it to update the model and generate updated diagram and code. It also generate Liquibase xml changeset. Moreover, it can also generate my GraphQL schema .. and update it when I change the data model.
This is the friggin' holy grail of all diagramming and modeling tools. And it is available and working now. Everything else is just too primitive !
Thanks! I was writing a FastAPI/SQLAlchemy+SQLModel/Rocketry app this week and dreaded documenting my model relationships. That said, I see a few SQLAlchemy extensions and Pydantic→Mermaid projects that can probably be used in my case, and that was the route I was initially planning on going down despite none of them being aesthetically pleasing.
Many of these online tools reasonably rely on human input to place tables/nodes. Do any of these recommended tools do a decent job of smartly laying out the final diagram?
Ah, that's nice to hear. This is one of my frustrations with Mermaid - trying to work out he magical correct ordering of entities to render something that isn't too chaotic.
Ah, diagrams from text. Every few months I look into producing slides, and look for something that saves me from having to use a mouse-intensive tool like PowerPoint or KeyNote, preferably using Markdown. There are plenty of tools to make markdown slides — slidev, Marp (esp w VSCode extension), HackMD, etc. But for diagrams I haven’t converged on a good workflow. Maybe mermaid is worth spending time with.
I love tools like this (text to diagram) and tend to use mermaid for most things. I also really like draw.io which includes a gui and has integration with GitHub to save diagrams directly to a source repo.
Is there a solution that has direct, bi-directional integration with Postgres/MySQL. For example, execute DDL from the diagram or point to an existing db and output the diagram text?
2. You should really work on error handling. I went to /app and tried changing the name of the CustomerID column, and nothing happened. I tried changing something else and it didn’t work either. It should tell me that the other table is mentioning a non-existent column and the input is invalid.
1. Every database has its own flavor of SQL syntax. So we need to settle on a minimal syntax that can export to any database. (BigQuery, Snowflake, MySQL etc..)
2. I tested it out. When i attempt to change CustomerID it does show a error toast in the bottom saying FK reference is invalid. Not sure, why its not showing up for you. Which browser are you using ?
I'm using Firefox. On my larger screen, the toast did appear, but it was outside of the scrolled-in view. On my smaller screen, the layout just broke and the toast is not visible.
When I see things like this, now my immediate question is always: what's the underlying model? Is this like a generally open and free thing, with maybe a little bit of monetization to keep the lights on? Then that's great.
But, if this is looking to be a big moneymaker? eh, nah.
Everyone launching a beta version of their product says its free. You invest time learning and migrating things over and by the time they catch up on features, pricing page shows up.
I have been using https://www.erdlab.io for 6 months now and so far it looks like they are true to their word. The developers on ERDLab have fulltime jobs and treat it as a side project but still push out updates and handle maintenance in a timely manner. I want them to succeed but not so much where they start to think about making money lol
OP Here. I can understand. I would be skeptical too
I have been using this tool just for myself for the last few months. The project is intentionally built for longevity (striving to keep the project for decades, if there are serious users). The costs to run are negligible, and I haven't thought much about monetizing. At this point, evaluating if there is enough interest in the tool
Is it possible to resize a table in the diagram? I work on a database that has a column name of "order_module_order_num" and if you try putting that in the text doesn't fit. The text runs into the data type.
Maybe an auto-complete or auto correct feature, or even syntax highlighting? I just found myself exploring for the right way to type and jumped to the solution of fuzziness which i agree, might not be ideal.
Suggest checking out the sqlparse library for a way to do the different flavours without needing to address each case directly: https://github.com/andialbrecht/sqlparse
neato! anyone interested in a collab on a more generic natural language graph making tool (think GoJS powered by LLM to make SysML diagrams etc). Demo of api here, I cant do front end, it is a weakness…
lpt: ask chatgpt or other sufficient LLMs for catchy domain name variations. I specifically ask for "in web 2.0 style".
I crank out short catchy brand names and domain names at far greater efficacy now, steers my thoughts really well if I’m not outright taking the first suggestion verbatim
> Generate DDL directly generate SQL statements to create your database tables.
Looks like it's capable of turning a diagram to SQL too? That's neat!
For those that are interested in an GPT-powered tool, we have a project that can generate diagrams from code of any language or natural language: https://eraser.io/diagramgpt
