AI Assistant
Rostyman includes a built-in AI assistant that can generate requests, write test scripts, explain responses, generate mock data, produce API documentation, and more. The assistant is context-aware: it knows which request tab you have open and what the last response looked like.
Opening the AI Panel
- Click the Sparkles (AI) icon in the right panel tab bar
- Press Ctrl+J to toggle the right panel, then click the AI tab
- The panel opens embedded in the right sidebar with a resizable width (280–600 px)
- The panel can be pinned (always visible) or unpinned (auto-hides and reopens on hover)
Setup
The AI assistant requires an API key from your chosen provider. Keys are stored locally in your settings database and are never sent to Rostyman's servers.
- Open the AI panel
- Click the Settings (gear) icon in the panel header
- Select your provider and enter the API key in the Model Selector area
Supported Providers
Anthropic (Claude)
| Model | Group | Best for |
|---|---|---|
| Claude Opus 4 | Claude 4 | Most capable — complex reasoning and analysis |
| Claude Sonnet 4 | Claude 4 | Balanced — speed and intelligence |
| Claude Haiku 3.5 | Claude 4 | Fast and efficient — everyday tasks |
| Claude 3.5 Sonnet | Claude 3.5 | Balanced — previous generation |
| Claude 3.5 Haiku | Claude 3.5 | Fast — previous generation |
OpenAI
| Model | Group | Best for |
|---|---|---|
| GPT-4.1 | GPT-4.1 | Latest flagship — best performance |
| GPT-4.1 Mini | GPT-4.1 | Lightweight — cost-efficient |
| GPT-4.1 Nano | GPT-4.1 | Fastest and cheapest — high-volume tasks |
| GPT-4o | GPT-4o | Multimodal — vision, audio, and text |
| GPT-4o Mini | GPT-4o | Affordable multimodal |
| o3 | Reasoning | Advanced reasoning — hard problems |
| o4-mini | Reasoning | Fast reasoning — cost-efficient |
Ollama (Local Models)
Rostyman supports Ollama for fully local, private AI inference with no API key needed:
- Install and run Ollama:
ollama serve - Pull a model:
ollama pull llama3.2(or any supported model) - In the AI panel, select Ollama as the provider
- The model dropdown automatically lists all locally available models (click the refresh icon to update the list)
Ollama connects to http://localhost:11434. No API key is required.
Configuration Panel
Click the Settings icon in the AI panel header to open the configuration panel.
Model Selector
A grouped dropdown picker showing models organized by family (Claude 4, GPT-4.1, Reasoning, etc.). Each model entry shows a one-line hint describing its strengths. Use the refresh icon to refetch the available models from your provider's API.
Streaming Toggle
When enabled, responses appear word-by-word as they are generated (streaming mode). When disabled, the full response appears after the AI finishes generating. Streaming is on by default and is supported by all three providers.
Chat Interface
The main chat view shows the full conversation history with the current AI conversation.
Sending Messages
Type your message in the input box at the bottom and press Enter (or click the Send button). Use Shift+Enter for a newline within your message.
Tip: Paste a curl command directly into the input box. Rostyman detects it and automatically fills in the method, URL, headers, and body of the current request tab.
New Chat
Click the + icon in the panel header to start a fresh conversation. The previous conversation is saved and accessible from the conversation list.
Conversation List
Click the List icon in the panel header to see all past conversations:
- Each conversation shows its title (auto-generated from the first message), provider, model, and timestamp
- Click any conversation to switch to it and resume from where you left off
- Hover to reveal a Rename or Delete button
AI Console
Click the Terminal icon in the chat view toolbar to toggle the AI console — a log of every API call made during the session. Each entry shows:
- Time, provider, and model
- Token count and duration
- Full request body (expandable)
- Full response (expandable)
- Error details if the call failed (auto-expanded)
The console is useful for debugging unexpected AI behavior or inspecting token usage.
Quick Action Buttons
Below the input box, a row of quick-action buttons trigger contextual AI prompts based on your active request tab. Buttons are enabled or disabled depending on whether a response is available.
| Button | Icon | When Available | What it Does |
|---|---|---|---|
| Generate | Wand | Always | Generates a request from plain English description |
| Explain | FileText | After sending a request | Explains the response in plain English |
| Tests | Sparkles | After sending a request | Writes Rostyman test scripts (rm.test() / rm.expect()) for the response |
| Fix | Bug | After a 4xx/5xx response | Analyzes the error and suggests fixes |
| Mock Data | Database | With a request tab open | Generates a realistic sample JSON request body for the endpoint |
| Document | BookOpen | With a request tab open | Generates concise Markdown API documentation for the endpoint |
| Validate Schema | Shield | With a request tab open | Validates the response JSON structure and reports added, missing, or type-mismatched fields |
Generate
Click Generate and describe what you want in natural language:
"POST request to create a new user with name, email, and role fields as JSON body, using Bearer token auth from the
{{api_token}}variable"
The AI fills in the method, URL (if mentioned), headers, body, and auth on the active request tab.
Explain
Sends the current response's status code and body (up to 2000 characters) to the AI and asks for a plain-English explanation. Useful for unfamiliar APIs, complex nested responses, or error messages you don't recognize.
Tests
Sends the current response's status code and body (up to 1000 characters) to the AI and generates relevant test assertions using Rostyman's scripting API:
rm.test("Status is 200", function() {
rm.expect(rm.response.code).to.equal(200);
});
rm.test("Response has id field", function() {
const body = rm.response.json();
rm.expect(body).to.have.property("id");
});
The generated scripts are placed directly into the chat — copy them to the Tests tab of your request.
Fix
Available when the last response was a 4xx or 5xx status. Sends the request method, URL, status, and response body to the AI, which suggests what might be wrong and how to fix it (missing headers, malformed body, auth issues, etc.).
Mock Data
Asks the AI to generate a realistic sample JSON request body for the current endpoint based on its method and URL. Useful when testing endpoints that require complex request bodies.
Document
Generates Markdown documentation for the current endpoint including:
- Endpoint description
- Parameters and headers
- Request body format
- Expected response structure
Validate Schema
Sends the response body to the AI and asks it to compare the JSON structure against what would be expected for the endpoint. Returns a report of:
- Matched fields
- Missing fields
- Extra/unexpected fields
- Type mismatches
MCP Tool Integration
When Rostyman is connected to external MCP servers (via the Servers tab in the MCP panel), those servers' tools become available to the AI assistant. The AI can call them automatically when answering questions — for example, asking the AI "look up issue #42 in GitHub" will invoke the GitHub MCP server's tools without you needing to make the API call manually.
Privacy
| Data | Where it goes |
|---|---|
| API keys | Stored locally in {userData}/rostyman.db (Rostyman settings table) — never sent anywhere |
| Request data (URL, headers, body) | Sent to your AI provider's API when using AI features |
| Response data | Sent to your AI provider's API when using Explain, Tests, Fix, or Validate |
| Conversation history | Stored locally in the settings database |
| Data to Rostyman servers | None |
Use Ollama for fully private AI inference where no data leaves your machine.