One Capriole AI account can connect Codex, Claude Code, Kilo Code, GitHub Copilot CLI, OpenCode, OpenClaw, Cline, Qwen Code, Crush, and Pi. Premium costs USD 9 a month for unlimited web chat with GPT-6.2 Astra, Claude Fable 5.1, Claude Opus 5.5, and Gemini 4 Carbon, and the API costs up to 92% less than official API prices, with 5M tokens a month included.
Each client keeps the wire format it expects. Capriole maintains OpenAI Responses, OpenAI-compatible Chat Completions, and Anthropic Messages paths for the documented integrations.
“OpenAI compatible” alone does not describe modern coding-agent support.
Why the maintained paths matter
Capriole removes the repeated provider setup behind these clients. One membership supplies the account, API balance, key management, latest-model aliases, and exact client-specific configuration. You do not need a separate provider account, key, balance, and billing page for each supported model family.
The result is a documented path from the browser workspace to repository work. Chat stays unlimited under Premium, while coding-agent traffic uses the visible 5 million monthly charged-token balance.
10 agents, several protocols
Coding agents may look similar at the prompt, but their model clients are not interchangeable.
The same account and API balance sit behind these integrations. The request shape still follows the client.
Codex sends Responses requests. Claude Code sends Anthropic Messages requests and can use token-counting preflight when the selected provider supports it. Cline and Crush use Chat Completions for every model. Kilo Code, OpenCode, Qwen Code, and Pi send each model family to the protocol Capriole serves for it: Responses for GPT, Messages for Claude, and Chat Completions for the other models. GitHub Copilot CLI and OpenClaw use a separate protocol path or provider for each model family.
Capriole preserves those distinctions. Its public Responses route is not secretly converted into Chat Completions, and its Messages route is not rewritten into an OpenAI response.
Why one key is useful
Without a shared access layer, trying several coding agents often means repeating the same administrative work:
- create or maintain accounts with different model providers;
- add billing details and balances in several places;
- store several API keys;
- learn which model name and base URL each client expects;
- revisit every configuration when a preferred model changes.
Capriole reduces that setup to one account and one Capriole API key. The integration guides provide the exact base URL, model selector, and client-specific fields.
The agents keep their own interfaces, tool systems, permission models, and local execution behavior. Capriole supplies model access; each coding agent still decides how to edit files, run tools, and present the result.
Latest aliases reduce maintenance
Model releases move quickly. A configuration tied to one concrete model ID is useful for reproducible work, but it also needs manual updates when you want the newest supported default.
Capriole provides maintained aliases for supported public paths:
openai-latest for the maintained OpenAI route;
claude-latest for the maintained Claude route;
google-latest on compatible public chat paths.
The alias is resolved before upstream dispatch and usage accounting. The coding agent sends the stable selector; Capriole records the resolved model that handled the request.
Concrete model IDs remain useful when an exact version matters for an audit, a test, or a long-running project.
What has been tested
The maintained setup follows actual client paths and tests.
Local validation covers all 10 clients through their documented configurations. Tests include basic prompts and supported tool-interaction flows.
Match each client to its protocol. Codex uses OpenAI Responses, Claude Code uses Anthropic Messages, and Cline and Crush use Chat Completions. The table above lists the configuration for each client.
The integration guides linked in the table are the source of truth for current fields and supported selectors.
Read how Capriole tests multi-model API compatibility for the difference between route coverage and a real coding-agent pass.
Browser chat and coding agents solve different parts of the job
Research, drafting, files, web search, and mid-conversation model changes can stay in the browser workspace. When the task needs repository context, local tools, or file changes, a coding agent can use the same Capriole account.
Capriole connects both surfaces to the same account. A Premium membership currently costs USD 9 per month, includes unlimited browser chat across the supported Premium model set, and adds 5 million charged API tokens each month. Coding agents use your Capriole AI API key and draw on your API token balance.
The plan keeps the two workloads separate. Everyday browser chat is unlimited under Premium, while programmatic and agent workloads use visible charged API tokens with top-ups available when needed.
Choose the agent first, then follow its protocol
Use Codex when its Responses-based OpenAI workflow fits the job. Use Claude Code when you want its Anthropic-native client. Kilo Code and Cline provide IDE or terminal workflows. GitHub Copilot CLI, OpenCode, OpenClaw, Qwen Code, Crush, and Pi provide more client choices.
Each documented client gets a maintained route to the models supported by that protocol.
Try GPT-6.2 Astra, Claude Fable 5.1, Claude Opus 5.5, or Gemini 4 Carbon in free browser chat, then upgrade when you are ready to create an API key. Start with the one-key coding-agent guide for the shared workflow, then open the integration page for the agent you use. Last modified on November 18, 2026