You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 6645959
Browse filesBrowse the repository at this point in the historyBrowse files
Generate API flow orchestration from natural language requirements — no manual node dragging needed.
325
+
Use natural language to let the LLM auto-complete flow orchestration, API integration and report design.
298
326
299
-
### How to Use
327
+
### Model Settings
300
328
301
-
1. Open the **Flow Designer** and click the "AI Generate" button in the toolbar
302
-
2. Expand "Model Settings" and fill in an OpenAI-compatible API base URL, API key and model name (DeepSeek / Qwen / Kimi / OpenAI etc.; saved in system config)
303
-
3. Describe your requirement, e.g.:
304
-
> Take a userId, call the get-user-info API, then query the user's order list, finally return the user name and the order list
305
-
4. Click "Generate & Replace Canvas" — the LLM picks APIs from the available list, auto-maps inputs (flow input params) and outputs (`env_` variables), wires nodes automatically, and supports all 14 node types including CONDITION branches and MYSQL (SQL) nodes
306
-
5. Fine-tune the result on the canvas, then save/deploy to expose it as an API
329
+
System Settings → AI Models: connect any **OpenAI-compatible API** (DeepSeek / Qwen / Kimi / OpenAI / GLM / Ollama etc.):
330
+
331
+
- Preset provider dropdown (or type your own) with auto-filled default base URL
332
+
- "Fetch models / Test" verifies the API key and pulls the available model list
333
+
- Default model dropdown + available models multi-select (switchable during chats)
334
+
- Multiple providers with enable/disable
335
+
336
+
### AI Assistants
337
+
338
+
The "AI Assistants" menu provides four assistants:
339
+
340
+
| Assistant | Capability |
341
+
|-----------|------------|
342
+
|**Assistant Management**| Custom assistants: name, system prompt, input params (text/number/date/switch/select), output params, quick-prompt buttons; auto-added to menu when enabled |
343
+
|**Flow Orchestration Assistant**| Describe requirements → auto-picks existing APIs, generates flow + input/output param definitions, condition branches, DB nodes (14 types) → preview, then confirm to create the flow |
344
+
|**API Integration Assistant**| Describe integration needs → generates suites & APIs (input / Header / output param configs) → preview, then confirm to import (deduplicated by code) |
345
+
|**Report Assistant**| Describe report needs → auto-generates datasets (reuses data views / data sources / flows / APIs), query params, layout (aggregates & formulas) → preview, then confirm to create the report |
346
+
347
+
Custom assistants support **multi-turn conversations**: chat bubbles, full history context, end-conversation to produce final JSON results from output params, new conversation and history viewing.
348
+
349
+
### In-Designer AI Generation
350
+
351
+
The Flow Designer toolbar also has an "AI Generate" button, e.g.:
352
+
> Take a userId, call the get-user-info API, then query the user's order list, finally return the user name and the order list
353
+
354
+
The LLM picks APIs, auto-maps inputs (flow input params) and outputs (`env_` variables), wires nodes with condition branches; confirm to apply, then fine-tune on the canvas and save/deploy.
- ✅ **🤖 AI orchestration** — Describe requirements in natural language and let the LLM auto-generate flow orchestration (OpenAI-compatible API; DeepSeek/Qwen/Kimi supported)
0 commit comments