One Coding Budget, Three API Protocols: A Practical Setup for Claude Code, Codex, and OpenCode

If you already use Claude Code, Codex, OpenCode, Kimi Code, or OpenClaw and want one metered budget instead of several disconnected model accounts, this guide shows how to choose the right protocol, configure the minimum required credentials, and verify the connection before trusting it with a repository.

AceDataCloud Coding Plan is currently a 30-day, Credit-metered multi-model API plan, not an unlimited editor subscription. The entry tier has a ¥99 display anchor and currently includes 500 Credits (400 base plus a 100-Credit promotion through November 30, 2026). Every tier has the same Coding model access; the tiers differ mainly in Credit allowance. Check the current Coding Plan page before buying because prices, promotions, and the live model catalog can change.

Start with the protocol, not the model name

The easiest configuration mistake is treating “client,” “protocol,” and “model” as the same choice. They are separate:

Do not assume that every model works through every client. Coding Plan currently exposes Claude Messages, OpenAI Responses, and OpenAI-compatible Chat Completions, but your Coding Application also has an API scope. A valid model must satisfy three conditions:

1. It appears in the current model catalog.
2. Your credential is allowed to call the selected API.
3. Your client understands that API protocol.

That separation matters when assigning work. You might keep Claude Code for a cross-file refactor, use Codex for an implementation-and-validation loop, and route repetitive test or documentation tasks through a configurable client using a GLM, Kimi, or DeepSeek model. The useful feature is not that one model is universally best; it is that the client does not have to dictate your entire model budget.

Minimum setup

Create a Coding Application after purchasing a plan, then copy its dedicated API token. Do not use a general account key and do not commit the token to your repository.

Use environment variables or your operating system's secret store:

export ACE_API_TOKEN="YOUR_CODING_APPLICATION_TOKEN" export ACE_BASE_URL="https://api.acedata.cloud" 

First, inspect the live model catalog:

curl "$ACE_BASE_URL/v1/models" \ -H "Authorization: Bearer $ACE_API_TOKEN" 

The id value of each returned item is the model string you pass in a request. It is not an Application UUID or service ID. The catalog is dynamic, and the global /v1/models response may be broader than your credential's allowed scope, so catalog presence alone is not proof of access.

For an OpenAI-compatible client, the essential settings are:

Base URL: https://api.acedata.cloud API key: YOUR_CODING_APPLICATION_TOKEN Model: <an id returned by /v1/models> 

The exact field names vary by client. Follow the Coding Plan integration center for client-specific paths rather than copying settings from an unrelated tool.

Verify before opening a large repository

Use this three-step check to avoid spending Credits on a broken agent loop.

1. Confirm authentication

A successful /v1/models response confirms that the token reaches the API. A 401 usually means the token is missing or invalid. If the token works in a shell but not in your editor, confirm that the editor process received the same environment variables, then restart it completely.

2. Confirm protocol and scope

Send the smallest meaningful request through the protocol your client will use. For an OpenAI-compatible Chat Completions client, the shape is:

curl "$ACE_BASE_URL/v1/chat/completions" \ -H "Authorization: Bearer $ACE_API_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "model": "MODEL_ID_FROM_THE_CATALOG", "messages": [{"role": "user", "content": "Reply with only: ready"}], "max_tokens": 8 }' 

This is a configuration example, not a benchmark. Replace the placeholder with a catalog ID compatible with your credential and protocol. A scope error means you should check the Coding Application's allowed APIs, not repeatedly rotate model names.

3. Run a bounded coding task

Start with a task whose output you can validate locally:

• explain one module and identify its public interface;
• add tests for one function without changing production code;
• review one diff and list only reproducible issues;
• perform a small refactor, then run the existing test command.

Avoid beginning with “analyze the entire monorepo.” Long context, tool calls, retries, parallel agents, uncached input, output, cache writes, and cache reads can all affect Credit consumption.

Choose a tier from measured work

The entry plan is useful for short projects, test coverage, and occasional use. Current public package data lists four 30-day tiers: Coding Plus (500 promotional Credits), Standard (1,600), Pro (4,000), and Max (10,000). These totals include limited-time bonuses through November 30, 2026; the base amounts are 400, 1,200, 3,000, and 8,000 Credits respectively.

A sensible purchasing loop is:

1. Pick one client and one real, bounded task.
2. Record Credits consumed for three representative runs.
3. Separate interactive work from unattended agent loops.
4. Add a buffer for retries and larger contexts.
5. Select the smallest tier that covers that measured workload.

Renewing before expiry carries remaining Credits forward and extends the term by 30 days. If the plan expires first, unused Credits expire. The plan does not auto-renew and does not fall back to a general balance, so set your own renewal reminder if continuity matters.

The practical value is control: keep the coding clients your team already understands, select models by task, and put the experiment behind a finite 30-day Credit budget. Start with protocol compatibility, validate with a tiny request, and let observed usage—not a guessed prompt count—determine the next tier.

References

• AceDataCloud Coding Plan
• Coding Plan integration guide
• Is the ¥99 Coding Plan Right for You?
• One Coding Plan for Claude Code, Codex, Kimi Code and OpenClaw

Product and configuration details are sourced from Ace Data Cloud's public plan data and documentation.