LLM API Request Builder
Build native OpenAI, Anthropic, and Gemini request JSON, cURL, JavaScript, and Python examples without API keys or model calls.
Native provider templates · Deterministic output
No key input · No body persistenceLLM API request workspace
Only generates request examples; never calls the model.
Chat Completions keeps system messages in messages, uses max_tokens, and enables streaming with stream: true.
Request parameters
OpenAICredentials are read only from an environment variable: OPENAI_API_KEY
Request preview
Read-only text; this page has no sending implementation.
{
"model": "gpt-4.1-mini",
"messages": [
{
"role": "system",
"content": "You are a concise, helpful assistant."
},
{
"role": "user",
"content": "Explain request builders in one sentence."
}
],
"temperature": 1,
"max_tokens": 1024,
"stream": false
}Portable configuration
Versioned configuration fields only; credentials, passwords, and unknown properties are rejected recursively. Failed imports never replace the current editor.
or drop a local file here (up to 0.5 MB)
Made for the task at hand.
Assemble reviewable API requests against curated native provider templates instead of assuming every model shares one schema. Edit ordered system, user, and assistant messages, inspect provider-specific transformations, and copy or download inert examples without entering a credential or contacting a model.
How to use
- Choose OpenAI, Anthropic, or Gemini, then edit the representative model ID, temperature, maximum output tokens, and streaming option.
- Add, edit, delete, and reorder messages. Review validation and mapping notes, especially Anthropic top-level system text and Gemini model roles and stream endpoints.
- Copy or download Request JSON, cURL, JavaScript, or Python. Export or import the allowlisted configuration separately; credentials are environment-variable placeholders only.
Your work stays yours.
Tool inputs are processed locally in your browser and are never sent to KitVerse servers. Clear the workspace when using a shared device.
Tool inputs are processed locally in your browser and are never sent to KitVerse servers. Clear the workspace when using a shared device.
Common questions
Does this tool call OpenAI, Anthropic, Gemini, or a backend?
No. It only renders request examples as text. There is no request-sending implementation, API key input, backend proxy, or model call. Code containing fetch or urllib is generated only for you to inspect, copy, or download.
How do provider mappings differ?
OpenAI keeps system in messages and uses max_tokens. Anthropic moves system to a top-level field, requires max_tokens and anthropic-version, and combines consecutive same-role turns. Gemini uses contents with user/model roles, system_instruction, and generationConfig.maxOutputTokens; streaming selects streamGenerateContent?alt=sse instead of adding a body flag.
Can an imported configuration contain an API key?
No. Imports are size-limited and allowlisted, and apiKey, authorization, x-api-key, password, and secret properties are rejected recursively. maxTokens and max_tokens remain valid token settings. Invalid imports leave the current editor unchanged.
Are messages saved automatically?
No. Message bodies and settings stay in page memory and disappear on refresh. Export or copy configuration explicitly if you want a local file or clipboard copy; neither contains real credentials.