Help · Documentation

Set Up deepllm With Coding Agents

Choose Codex CLI or Claude Code, install the client, connect it with an API token from the API Access page, then run it inside a real project folder.

Claude Code

ANTHROPIC-COMPATIBLE

Anthropic's coding agent through the deepllm gateway.

OpenAI-compatible base URL

https://www.deepllm.dev/v1

Anthropic-compatible base URL

https://www.deepllm.dev

Claude Code — Windows

01 — Install Claude Code

Recommended method: native PowerShell installer. Alternative: WinGet.

powershell
1irm https://claude.ai/install.ps1 | iex
2# or
3winget install Anthropic.ClaudeCode

Install Git for Windows if you want Claude Code to use Git Bash. Without it, Claude Code can still use PowerShell.

02 — Create your deepllm token

Open API Access in your dashboard and press Create token. Copy the token — it starts with al_ — and use it wherever the steps below say YOUR_AEROLINK_API_KEY.

Treat the token like a password: anyone who has it can spend your balance. If it leaks, revoke it on the API Access page and create a new one.

03 — Create settings.json

Save this file at C:\Users\YOUR_USERNAME.claude\settings.json.

powershell
New-Item -ItemType Directory "$env:USERPROFILE\.claude" -Force
json
1{
2  "env": {
3    "ANTHROPIC_API_KEY": "YOUR_AEROLINK_API_KEY",
4    "ANTHROPIC_BASE_URL": "https://www.deepllm.dev",
5    "ANTHROPIC_MODEL": "base-claude",
6    "ANTHROPIC_DEFAULT_SONNET_MODEL": "base-claude",
7    "ANTHROPIC_DEFAULT_OPUS_MODEL": "base-claude",
8    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "base-claude",
9    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
10  },
11  "permissions": {
12    "allow": [],
13    "deny": []
14  },
15  "apiKeyHelper": "echo 'YOUR_AEROLINK_API_KEY'"
16}

Replace both YOUR_AEROLINK_API_KEY values with your token. Claude Code sends every request to the deepllm gateway, which serves the model named in ANTHROPIC_MODEL — pick it from the list on the API Access page. The three ANTHROPIC_DEFAULT_* lines point Claude Code's built-in Sonnet, Opus and Haiku choices at that same model, so switching with /model inside Claude Code does not change which model you are billed for.

04 — Check your token works

Run this before configuring any tool. A list of models means your token and connection are working — every id in it is a model name you can use.

powershell
curl.exe https://www.deepllm.dev/v1/models -H "Authorization: Bearer YOUR_AEROLINK_API_KEY"

05 — Verify and start

bash
1claude --version
2claude doctor
3cd path-to-your-project
4claude

Troubleshooting

ErrorWhat it meansWhat to do
401 Invalid or missing tokenThe token is wrong, revoked, or not loaded in this terminalCopy it again from API Access. If you saved it as an environment variable, open a new terminal.
402 Balance exhaustedYour balance can't cover the requestRecharge from your dashboard.
403 Account not activeYour account is suspended or still waiting for activationContact us through the store.
404 Model not availableThat model name is not on your planUse a model id from the API Access page (or from /v1/models).
429 Too many requestsYou hit your plan's per-minute or parallel-request limitWait a few seconds. Agents retry automatically.
5xx Upstream errorThe model provider failedRetry after a few seconds.

Every response carries an x-request-id header (it starts with req_). Include it when you contact support — it lets us find your exact request.

FAQ

Which model should I use?

The API Access page lists every model available on your plan with its price per million tokens. base-claude is used in this guide as an example; replace it with any model from that list.

Why does /model not change what I'm billed for?

The three ANTHROPIC_DEFAULT_* lines map Claude Code's built-in Sonnet, Opus and Haiku choices to the model you set. To use a different model, change all four base-claude values in settings.json and restart Claude Code.

Which tool should I pick?

Every tool works with the same deepllm token and balance, so pick the one you like. Claude Code uses the Anthropic Messages API (https://www.deepllm.dev). Codex, OpenCode and most other agents use the OpenAI Chat Completions API (https://www.deepllm.dev/v1). The More agents guide covers Cline, Roo Code, Kilo Code, Continue, Aider, Qwen Code, Crush and Zed.

Why does the guide mention Git?

Git is not required for every command, but real projects use it for history, undo, branches, collaboration, and deployment. Agents also use Git status and diffs to avoid mixing unrelated work.

Where does my usage show up?

Every request is metered by the deepllm gateway. Activity Logs shows each request with its model, tokens and cost, and your balance history on the dashboard lists every charge.