CodeWhale (formerly DeepSeek-TUI) Integration

Connect CodeWhale to QCode.cc: Claude via the anthropic provider, GPT and Chinese models via the OpenAI-compatible provider

Last verified: 2026-09-18 · 📄 Per official docs (CodeWhale v0.9.13, released 2026-09-14 (formerly DeepSeek-TUI))

At a glance¶

⚠️ The project was renamed. DeepSeek-TUI is now CodeWhale. The binary changed from deepseek to codewhale, and the config file moved from ~/.deepseek/config.toml to ~/.codewhale/config.toml. The fallback is conditional: the official migration contract is read-with-fallback, write-to-new — reads look in ~/.codewhale/ first and fall back to ~/.deepseek/ only when that is the only directory present, while every write goes to the new directory. Also, since v0.9.0 the deepseek and deepseek-tui commands were deleted; the entry points are codewhale, codew (a convenience alias) and codewhale-tui. The old site deepseek-tui.com now 301-redirects to codewhale.net. This page's URL is unchanged.

CodeWhale is a terminal AI coding agent with dozens of first-class providers (anthropic, openai, deepseek, ollama, vllm, openrouter and more; the docs point at ProviderKind::ALL in the source as the real list, which grows and shrinks between versions — run /provider to see the set in your own build). It speaks both the native Anthropic Messages protocol and the OpenAI Chat Completions protocol, and either one can point at QCode.cc.

🔴 Read this first: Claude requires the anthropic provider¶

QCode's OpenAI-compatible endpoint does not accept Claude models — putting claude-… into [providers.openai] returns model_not_available_on_endpoint. See the compatibility table in Endpoints & API Paths.

Why use CodeWhale with QCode¶

• Familiar TUI: Plan / Work / Operate modes, with the Ask / Auto-Review / Full Access permission postures (Shift+Tab) and built-in MCP / Shell / Git / subagents
• One API key: shares your QCode plan quota with Claude Code and Codex CLI
• Provider switching: move between anthropic / openai / ollama / vllm in the same tool
• Works well from mainland China: asia.qcode.cc (Asia node, near HK/JP) has the lowest latency
• Fully open source: MIT licensed, the config file is auditable

1. Install¶

The official first choice is the GitHub Releases install script; npm and Cargo are called secondary packaging routes in the docs:

# Officially recommended (macOS / Linux): installs the release binary
curl -fsSL https://codewhale.net/install.sh | sh

# npm - the official docs call this a secondary packaging route (Node 18+)
npm install -g codewhale

# Homebrew - add the tap first, otherwise a fresh machine cannot find the formula
brew tap Hmbown/deepseek-tui
brew install codewhale

# Cargo - source build; the crate is codewhale-cli, the command it installs is codewhale
cargo install codewhale-cli --locked

Platform limits and details: official install docs.

Verify (trust the actual output, not a version number printed in any doc):

codewhale --version

2. Configure Claude (anthropic provider)¶

Edit ~/.codewhale/config.toml:

# ~/.codewhale/config.toml

provider = "anthropic"

[providers.anthropic]
api_key  = "cr_your_qcode_key"
base_url = "https://api.qcode.cc/api"
model    = "claude-sonnet-5"

From mainland China, swap the host for https://asia.qcode.cc/api (Asia node, near HK/JP). The key is unchanged.

Environment variables work instead of the config file; the docs now favour the generic CODEWHALE_* set:

export CODEWHALE_PROVIDER="anthropic"
export CODEWHALE_BASE_URL="https://api.qcode.cc/api"
export CODEWHALE_MODEL="claude-sonnet-5"
export ANTHROPIC_API_KEY="cr_your_qcode_key"

codewhale

The provider-specific variables (ANTHROPIC_BASE_URL / ANTHROPIC_MODEL) are still accepted, and so is the codewhale --provider anthropic flag.

3. Configure GPT and Chinese models (openai provider)¶

Multiple providers can coexist in one config; switch with codewhale --provider <id>:

[providers.openai]
api_key  = "cr_your_qcode_key"
base_url = "https://api.qcode.cc/openai/v1"
model    = "gpt-6-sol"

Stop base_url at /openai/v1; CodeWhale appends /chat/completions. If a build needs a different suffix, the official key for that is path_suffix inside [providers.openai] — do not smuggle it into base_url.

The four Chinese families (glm-5.2 / kimi-k3 / deepseek-v4-pro / qwen3.8-max …) work on both legs, so either provider is fine. Ids are listed in Chinese Models.

4. Available models¶

4.x models such as claude-sonnet-4-6 and claude-opus-4-8 are still sold. The full list and live prices are on qcode.cc/models.

List what you can call right now — note the two legs return different lists:

# Claude and Chinese models (Anthropic leg)
curl https://api.qcode.cc/v1/models -H "Authorization: Bearer cr_your_qcode_key"

# GPT (OpenAI leg)
curl https://api.qcode.cc/openai/v1/models -H "Authorization: Bearer cr_your_qcode_key"

5. Verify connectivity¶

KEY="cr_your_qcode_key"

# Claude (Anthropic protocol) — should return JSON containing content
curl -X POST https://api.qcode.cc/api/v1/messages \
  -H "x-api-key: $KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","max_tokens":32,"messages":[{"role":"user","content":"ping"}]}'

# GPT (OpenAI protocol) — should return JSON containing choices
curl -X POST https://api.qcode.cc/openai/v1/chat/completions \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-sol","max_tokens":32,"messages":[{"role":"user","content":"ping"}]}'

Then launch:

codewhale

6. Troubleshooting¶

Related¶

• Endpoints & API Paths — protocol × model-family table
• Chinese Models — GLM / Kimi / DeepSeek / Qwen ids and paths
• Model Selection — which model for which task