AI assistant¶

DiagramaHub follows a BYOL model (Bring Your Own LLM): you connect your own API key from an AI provider, and DiagramaHub uses it to generate, improve, fix and explain diagrams. Model usage is billed to your provider account, not by DiagramaHub.

Supported providers¶

Provider	Where to get the API key
Google Gemini	Google AI Studio
OpenAI GPT	OpenAI Platform
Anthropic Claude	Anthropic Console
DeepSeek	DeepSeek Platform
Minimax	Minimax Platform

Configure a provider¶

Go to Settings → AI Integrations.
Click Add Provider and choose the provider.
Paste your API Key and choose the Model. The suggested model is marked Recommended.
Optionally give it a Custom name (for example "My Gemini account").
Use Test Connection to check the key before saving it.

If the test fails, the message tells you why: invalid key or missing permissions, an account with no credits, the provider's rate limit, or a model that is not available for that key.

You can connect several providers. Mark one with Use as default and temporarily turn off the ones you don't want to use with Active. A connection's provider cannot be changed: if you want another one, create a new connection.

Available models (0.8.x)

The list comes from the server's model catalog. It may change between versions.

Provider	Models	Recommended
Google Gemini	`gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.1-pro-preview`	`gemini-3.8-flash`
OpenAI GPT	`gpt-6-luna`, `gpt-6.1-sol`, `gpt-6-astra`, `gpt-5.5`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-4.1-mini`	`gpt-6-luna`
Anthropic Claude	`claude-sonnet-5-5`, `claude-opus-5-5`, `claude-fable-5-1`, `claude-haiku-4-5-20251001`	`claude-sonnet-5-5`
DeepSeek	`deepseek-flash`, `deepseek-v4-pro`	`deepseek-flash`
Minimax	`MiniMax-M2.7`, `MiniMax-M3`, `MiniMax-M2.7-highspeed`	`MiniMax-M2.7`

If a model you already had configured leaves the list, it is shown as retired. It keeps working as long as the provider offers it, and you can change it when editing the connection.

Your key is protected

The key is stored encrypted in the database.
It is never returned in full: the UI only shows a masked version (for example AIza...xyz).
When editing a connection, leave the field empty to keep the current key.
The DiagramaHub server is what calls the provider. Your diagram's code and your messages are sent to the provider you chose, under its own privacy policies.

The AI chat¶

In the editor, open AI Chat. The chat is the main way to create and modify diagrams with AI: describe what you want in plain language and the assistant replies with the complete diagram.

sequenceDiagram
    participant Y as You
    participant D as DiagramaHub
    participant P as AI provider
    Y->>D: "Add a validation step before payment"
    D->>P: Instructions + current diagram + history
    P-->>D: Reply with the diagram
    D->>D: Validates the syntax (and asks for a fix if it fails)
    D-->>Y: Diagram preview
    Y->>D: ✓ Accept or ✗ Reject

Quick actions¶

Below the message box there are four shortcuts:

Action	What it asks the model
🧠 Explain	To explain the diagram clearly and briefly. It does not change the diagram.
🎨 Improve UI	To make it clearer, cleaner and easier to read.
⚙️ Improve Process	To improve the process it represents so it is more efficient.
🛠️ Fix	To fix syntax, structure or rendering errors and return the complete diagram.

Review the changes¶

When the assistant proposes a diagram, the chat shows a Diagram preview (with View larger):

✓ Accept applies the change to the editor.
✗ Reject discards it.
On earlier proposals you can use ↩ Restore or ↩ Use to go back to that version.

Chat sessions¶

Each diagram has its own chat sessions. You can start a New session, switch between sessions, rename or delete them, and pick each session's Model.

When a conversation gets close to the model's context limit, DiagramaHub summarizes it automatically and continues in a new session that starts from that summary. The previous session is marked Session finished (read-only): you can read it, but you can no longer send messages to it.

Language¶

The assistant replies in the UI language (Spanish or English), including the labels of the diagrams it generates.

Automatic fixing¶

With Fix generated diagrams automatically (in Settings → AI Integrations, on by default), if the Mermaid or D2 code the AI generates cannot be rendered, DiagramaHub asks the model for one fix before showing it to you:

Mermaid is validated in your browser with the same engine that draws the diagram.
D2 is validated on the server.
PlantUML and DBML are not fixed automatically: their validation produced false positives.

Each automatic fix uses one extra call on your key

If you prefer to control usage, turn this option off and use Fix only when you need it.

Fix with AI from the editor¶

If the diagram has an error, the code panel shows a Fix with AI button. The proposal is shown next to the original code so you can review the changes before applying them.

Replies that are too long¶

If the diagram the model generates is so large that the reply is cut off at the output limit, DiagramaHub does not apply that incomplete code. Instead, the chat tells you the reply was cut off. Ask for a smaller diagram or build it in parts.

Other AI features¶

AI description: in the diagram description, Generate with AI proposes a text. You can Refine it with instructions ("make it shorter", "more technical detail") before you Accept it.
Convert diagram type: converts the diagram to another format (for example from PlantUML to Mermaid). The conversion is shown in a preview with warnings about possible incompatibilities and is applied only when you click Apply conversion. More in The editor.

No provider configured¶

If you try to use an AI feature without an active provider, you will see AI Configuration Required with a Go to Configuration shortcut.