Cursor, Aider and other tools

Cursor, Cline, Continue, Aider, OpenCode and any tool that lets you change the OpenAI base URL work with heura. Set the base URL, paste your key, type a model name.

The three values every tool needs

Base URL https://api.useheura.com/v1. API key: your heura key. Model: a name from GET /v1/models, for example deepseek/deepseek-v4.1-flash.

ToolWhere to put it
CursorSettings, Models, override the OpenAI base URL, paste the key, add the model name as a custom model
Cline / Roo CodeProvider: OpenAI Compatible. Base URL, API key and Model ID fields
ContinueA model entry with provider openai, apiBase set to the base URL, and apiKey
AiderEnvironment: OPENAI_API_BASE and OPENAI_API_KEY, then aider --model openai/<model name>
OpenCodeA custom provider using the OpenAI-compatible adapter with the base URL and key
LiteLLM, LangChain, the OpenAI SDKsbase_url or baseURL set to the base URL
Tool menus change

The settings above describe where these tools put the base URL, and the tools move their menus. Only Claude Code, the official OpenAI and Anthropic SDKs were run against heura in tests. If a tool cannot take a custom base URL, it cannot use heura.

Plain curl

bash
curl https://api.useheura.com/v1/chat/completions \
  -H "Authorization: Bearer $HEURA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek/deepseek-v4.1-flash","max_tokens":200,
       "messages":[{"role":"user","content":"hello"}],"stream":true}'

Choosing a model

Heura routes by the exact model name. The names come from the catalog (GET /v1/catalog) and a model works only while at least one healthy seller offers it (GET /v1/models and the Market page). If a name has no seller you get a clear error, never a silent switch to a different model.

Streaming, tools and limits

  • Streaming and tool calls work on both the OpenAI and the Anthropic format.
  • Request bodies up to 8 MB. Each key is limited to 600 requests per minute.
  • Embeddings and the legacy completions path are passed to the gateway, but only chat and streaming have been tested end to end.