OpenAI compatibility

LLMBase is OpenAI-compatible for chat inference. In practice, that means most applications using the OpenAI SDK for chat.completions.create() can move to LLMBase by changing the base URL, API key, and model ID.

Migration checklist

From	Change
OpenAI	Set baseURL to https://api.llmbase.ai/v1, use an llmbase_... inference key, and choose an LLMBase model ID
Generic OpenAI-compatible client	Replace its base URL with https://api.llmbase.ai/v1, remove unsupported non-standard fields, and use an llmbase_... inference key

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.llmbase.ai/v1",
  apiKey: process.env.LLMBASE_API_KEY,
});

const response = await client.chat.completions.create({
  model: "<model-id-from-/v1/models>",
  messages: [{ role: "user", content: "Write a short deployment checklist." }],
});

console.log(response.choices[0]?.message.content);

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.llmbase.ai/v1",
    api_key=os.environ["LLMBASE_API_KEY"],
)

response = client.chat.completions.create(
    model="<model-id-from-/v1/models>",
    messages=[{"role": "user", "content": "Write a short deployment checklist."}],
)

print(response.choices[0].message.content)

Compatible endpoints

Endpoint	Purpose
POST /v1/chat/completions	Chat, streaming, tool calling, structured outputs, multimodal input on supported models
GET /v1/models	OpenAI SDK model discovery
GET /v1/models?metadata=true	Detailed model metadata with pricing, modalities, and capabilities
GET /v1/models?filter=agents&metadata=true	Detailed metadata filtered to agent-compatible models
GET /v1/balance	LLMBase prepaid inference balance

LLMBase focuses on chat inference for curated open-source models. It does not try to mirror every vendor-specific endpoint. Use the direct API for LLMBase-hosted chat inference with prepaid inference keys; use the LLMBase Chat app or agent API when you want chat-subscription agent features instead.

Supported request features

The direct inference API supports the portable chat features most production clients need:

Feature	Request field
Streaming	stream: true
Function tools	tools, tool_choice
JSON mode	response_format: { "type": "json_object" }
JSON Schema output	response_format: { "type": "json_schema", ... }
Log probabilities	logprobs, top_logprobs
Reasoning output	reasoning_effort, extra_body.chat_template_kwargs on reasoning-capable models
Prompt caching	prompt_cache_key
Sampling controls	temperature, top_p, top_k, min_p, repetition_penalty, penalties, stop, seed

Advanced features are model-dependent. When your app requires tools, response_format, reasoning output, or logprobs, inspect supported_features and supported_parameters from /v1/models?metadata=true before choosing a model.

Models that advertise reasoning can return reasoning_content in the OpenAI-compatible assistant message. Use reasoning_effort where supported for portable effort control. LLMBase also forwards the common template flags enable_thinking, thinking, and preserve_thinking from extra_body.chat_template_kwargs for model families that use those names.

Feature guides:

• Streaming for Server-Sent Events
• Tools for function calling
• Structured outputs for JSON responses
• Prompt caching for prompt_cache_key
• Reasoning for thinking traces and effort control

Differences from other OpenAI-compatible APIs

LLMBase requests choose one public model ID from GET /v1/models. Some non-standard fields accepted by other APIs are not part of the LLMBase direct inference contract.

Remove unsupported fields before sending direct LLMBase requests:

Field	LLMBase equivalent
provider	Choose the LLMBase model ID
models / route	Use one model ID
plugins	Use LLMBase-supported request fields such as tools or response_format
debug	Not part of the public direct inference contract
service_tier	Not part of the public direct inference contract
top-level cache_control	Use prompt_cache_key

If your application depends on non-standard marketplace controls, keep that integration separate. For LLMBase models with direct inference billing, use https://api.llmbase.ai/v1 and the documented request fields above.

Differences from native model APIs

Native model APIs often expose model-family-specific endpoints. LLMBase currently documents the direct chat inference surface, not every native vendor API.

Use LLMBase when you want:

• one curated model ID namespace
• prepaid inference keys
• OpenAI-compatible chat requests
• prompt-cache billing when responses report cached tokens
• dynamic model metadata with supported capabilities

Use a native model API when your workflow depends on model families outside LLMBase chat inference, such as custom image generation endpoints, native webhooks, private deployment controls, embeddings, reranking, or speech APIs.

Cost controls

For predictable spend:

1. Use GET /v1/balance before starting long prepaid jobs.
2. Pick a model from /v1/models?metadata=true and read its pricing.prompt, pricing.completion, and optional pricing.input_cache_read.
3. Set max_tokens for user-facing requests.
4. Use a stable prompt_cache_key for repeated long system prompts, agent conversations, or workspace-level prompts.
5. Keep large static instructions at the beginning of messages and changing user input later, so prompt caches can reuse the prefix.

Common mistakes

Symptom	Check
401 invalid_api_key	Use an llmbase_... inference key, not another service’s key, an OpenAI key, or a llmbase_chat_... key
402 insufficient_quota	Add prepaid inference credits, check the key spend cap, or use the Chat app/agent API with a Pro subscription
404 Model not found	Use an ID returned by GET /v1/models
400 invalid_request_error for tools or JSON	Choose a model that advertises the needed supported_features
SDK receives HTML instead of JSON	Verify the base URL and send Authorization: Bearer $LLMBASE_API_KEY on model discovery and completions requests
Request works elsewhere but not direct	Remove vendor-specific fields such as provider, models, route, and plugins