Coding Agents
Any coding agent that can talk to an OpenAI-compatible endpoint can use Flow AI. Every guide in this section uses the same three settings:
Before you start
- Give each tool its own key. Coding agents send long prompts and many requests. A key per tool, with a spending limit, keeps a runaway session from using your whole balance, and gives each tool its own line in Usage.
- Pick a model with tool calling. Agents read files, run commands and edit code through tool calls. A model without tool calling can chat but can’t act.
- Pick a large context window. Agent system prompts and tool definitions are long, and a session grows fast. Models with 128K tokens or more work best.
Flow AI serves the OpenAI chat completions API. Tools that only speak Anthropic’s Messages API (such as Claude Code) or OpenAI’s Responses API (such as Codex CLI) can’t use Flow AI yet.
Guides
Next: Cursor — use Flow AI models in Cursor.