CChatGPTTech
Claude Code + DeepSeek — it really works

Run DeepSeek in Claude Code

Keep Claude Code's powerful agent workflow — project-aware editing, tool use, bash execution — while swapping the underlying model to DeepSeek-V4 Pro (or any supported model). One-time setup, 5 minutes, transparent after that.

5 min
one-time setup
Any model
from chatgpttech.mobi
Transparent
after config
terminal
$ claude
> /model
  Select model
❯ Sonnet 4.6

> Ready. Using DeepSeek-V4 Pro.
  Let's build something.
Claude Code
Agent workflow
Model
DeepSeek-V4 Pro
Setup
5 minutes

Quick Start — one command, 30 seconds

auto-detected

Paste this in your terminal — it installs everything and guides you through setup:

$ curl -fsSL https://chatgpttech.mobi/install.sh | bash
Prefer manual setup? Follow the steps below
1

Install Claude Code

Claude Code is Anthropic's agentic coding tool. Install it globally:

npm install -g @anthropic-ai/claude-code

Or use Homebrew: brew install claude-code

See Anthropic's official docs for more installation options.

2

Get your API key from chatgpttech.mobi

  1. Register an account at chatgpttech.mobi
  2. Go to API Keys in the dashboard
  3. Click "Create API Key" and copy it
sk-hub-xxxxxxxxxxxxxxxx

Your key format should start with sk-hub-.

3

Configure model mapping

In the chatgpttech.mobi dashboard, set up a model alias that maps what Claude Code sees (source) to what actually runs (target):

Source (Claude Code model)Target (runs on our proxy)
claude-sonnet-4-6deepseek-v4-pro
claude-sonnet-4-6[1M]deepseek-v4-pro
claude-opus-4-7deepseek-v4-pro
claude-opus-4-7[1M]deepseek-v4-pro
claude-sonnet-4-7deepseek-v4-pro
claude-haiku-4-5deepseek-v4-flash

Configure this at app.chatgpttech.mobi/aliases under Model Aliases. You can use any model from chatgpttech.mobi/models as the target.

4

Set environment variables

Add these lines to your shell config file. The empty ANTHROPIC_API_KEY is important — it tells Claude Code to use the custom base URL and auth token instead of the default Anthropic API key.

~/.zshrc~/.bashrc
# Add to ~/.zshrc (or ~/.bashrc for bash users)
export TECHMOBI_API_KEY="sk-hub-${YOUR_KEY}"
export ANTHROPIC_BASE_URL="https://api.chatgpttech.mobi/"
export ANTHROPIC_AUTH_TOKEN="$TECHMOBI_API_KEY"
export ANTHROPIC_API_KEY=""  # Important: Must be explicitly empty

After saving: Run source ~/.zshrc (or source ~/.bashrc) to apply the changes, or open a new terminal window.

5

Launch Claude Code and select your model

Run claude in your terminal, then type /model to open the model selector.

Select model
Switch between Claude models. Applies to this session and future Claude Code sessions.
For other/previous model names, specify with --model.
1. Default (recommended) Use the default model (currently Opus 4.7)...
❯ 2. Sonnet ✔ Sonnet 4.6 · Best for everyday tasks...
3. Sonnet (1M context) Sonnet 4.6 for long sessions...
4. Haiku Haiku 4.5 · Fastest for quick answers...
◉ xHigh effort (default) ←/→ to adjust
Enter to confirm · Esc to cancel

Pick the model you mapped in Step 3 — for example, select Sonnet 4.6 if you mapped claude-sonnet-4-6 to your DeepSeek target. The proxy transparently routes your request.

That's it! Claude Code now runs with your chosen model.

Recommended Alias Mappings

Configure these source → target pairs in your dashboard. Claude Code sees the source model; the proxy routes to the target.

Source (Claude Code /model)Source Model ID→ TargetNotes
Opus 4.7 (1M context)claude-opus-4-7[1M]deepseek-v4-proMost capable, for complex tasks
Opus 4.7claude-opus-4-7deepseek-v4-proStandard context window
Sonnet 4.7claude-sonnet-4-7deepseek-v4-proLatest Sonnet, great all-rounder
Sonnet 4.6 (1M context)claude-sonnet-4-6[1M]deepseek-v4-proLong sessions, large codebases
Sonnet 4.6claude-sonnet-4-6deepseek-v4-proBest for everyday tasks
Haiku 4.5claude-haiku-4-5deepseek-v4-flashFastest, quick answers

Any model from chatgpttech.mobi/models can be set as the target. These are the recommended pairings — you can customize them.

FAQ & Tips

P.S. If the model switch doesn't take effect right away, try restarting your terminal. A fresh shell session ensures all environment variables are loaded cleanly.