Skip to content

Coding assistants

Coding assistants are often among the biggest AI costs in an engineering organisation, and the least visible: each developer brings a personal key or a subscription, and nobody sees which code and prompts went where. Point the assistant at the gateway instead and every call is checked against your policies, charged to the developer's team and recorded in Request logs. Give each developer a key of their own: their usage shows per person and per team, a limit on the key keeps an autonomous task's spend in bounds, and switching off one key ends one person's access without touching anyone else.

Assistant Format it speaks The field to set Custom endpoint
Claude Code Anthropic ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN Yes
GitHub Copilot OpenAI or Anthropic Custom Endpoint model in VS Code (url, apiKey) Chat and agent in VS Code only; completions stay with GitHub
Cursor OpenAI Override OpenAI Base URL and OpenAI API Key Chat only; calls come from Cursor's servers, Tab stays with Cursor
Cline OpenAI or Anthropic OpenAI Compatible: Base URL, API Key, Model Yes
Roo Code OpenAI OpenAI Compatible: Base URL, API Key, Model Yes
Continue OpenAI apiBase and apiKey in config.yaml Yes
Zed OpenAI or Anthropic api_url under openai_compatible or anthropic_compatible Yes (not edit predictions)
Aider OpenAI OPENAI_API_BASE and OPENAI_API_KEY Yes
Windsurf (Devin Desktop) — — No

Every assistant here that speaks the OpenAI format can use Claude and Gemini models as well as GPT, and Claude Code can use GPT or Gemini: the gateway translates the request and the answer both ways. Use the model names your organisation approved under Models in the Control room.

What you need for any of them: the gateway's address (under Connect your app on the Keys page, and in System › Settings › Gateway) and a Vidai key from Keys › Issue a key. The key starts vk_; never use a provider's key.