Langflow import
Cuttlely loads a Langflow JSON export only when the graph is a simple chat. Anything else stays on the import screen with “Langflow import is not supported yet.” The file is not opened as a partial canvas.
Accepted format
Section titled “Accepted format”A Langflow export is a JSON object whose data.nodes (or top-level nodes) are React Flow nodes. A node that counts has type genericNode or data.node.template. Chat Input, Chat Output, Text Input, Text Output, and notes are ignored. They are the Cuttlely chat box, not canvas nodes.
The graph must contain:
- exactly one chat model
- at most one Prompt
- no other component
An empty Language Model provider is read as OpenAI. The canvas note says so, so you can swap the chat model.
Supported providers: OpenAI, Anthropic, Google, Groq, Ollama, Mistral, Azure OpenAI, and xAI. Component types such as ChatOpenAI, OpenAIModelComponent, and LanguageModelComponent use that list.
Mapping
Section titled “Mapping”| Langflow | Cuttlely |
|---|---|
| Chat model | The matching chat-model node. model or model_name becomes Model Name. Temperature and a positive max token count are copied. |
| Prompt template | Conversation Chain system message. A {variable} fed by Chat Input or Text Input is removed, including a line that is only User: or Answer:. |
| Chat Input and Chat Output | Dropped. The Conversation Chain is the node you chat with. |
| (added) | Buffer Memory, because a Conversation Chain needs one. |
| API keys, passwords, and secret fields | Not copied. Select a credential saved in this app. |
The canvas also gets a sticky note that says the keys were left behind.
Out of scope
Section titled “Out of scope”Agents, tools, vector stores, retrievers, file loaders, more than one model, more than one prompt, and any component not listed above. A second model or an unknown provider is the same refusal. Workspace export and import stay off.
What are you building?
Section titled “What are you building?”Chatflows, Harnesses, Agentflows, ETL, and Templates open with outcome choices. An empty list keeps that card open. A list that already has items can hide it. Which should I use? opens the names:
- Agent team is a harness. The sidebar and the list say Harness · Agent Team: Harness Prime and specialists, called by name from a chatflow. Langflow has no harness.
- Chat / simple AI app is a Chatflow: one chatbot or a simple flow you chat with.
- Workflow / automation is an Agentflow.
- Knowledge / data pipeline is ETL.
- Import is a menu in the Chatflows header, and the canvas gear has it too. Bring flows from another tool under What are you building? links to Bring your existing flows. Flowise JSON opens in the migration wizard (
docs/flowise-migration-wizard.md). Langflow JSON uses the rule above. - Credentials stay in this app. Create one, then select it on the chat model.