Skip to content

deepseek.md

Latest commit

 

History

History
105 lines (73 loc) · 4.65 KB

deepseek.md

File metadata and controls

105 lines (73 loc) · 4.65 KB
Raw

Deepseek

📌 Important Note

Before configuring, please review:

Example Configuration

Basic Setup

aicommit2 config set DEEPSEEK.key="your-api-key"
aicommit2 config set DEEPSEEK.model="deepseek-v4-flash"

Advanced Setup

aicommit2 config set DEEPSEEK.key="your-api-key" \
    DEEPSEEK.model="deepseek-v4-flash" \
    DEEPSEEK.temperature=0.7 \
    DEEPSEEK.maxTokens=4000 \
    DEEPSEEK.locale="en" \
    DEEPSEEK.generate=3 \
    DEEPSEEK.topP=0.9

Settings

Setting Description Default
key API key -
model Model to use deepseek-v4-flash
thinking Enable thinking mode Auto (see below)
reasoningEffort high or max (only when thinking is enabled) high

aicommit2 does not maintain a strict allowlist of model names; unsupported names surface as API errors from DeepSeek.

Thinking mode

Use DEEPSEEK.thinking / DEEPSEEK.reasoningEffort in config; aicommit2 sends thinking and reasoning_effort in the JSON body (OpenAI-compatible), matching what the Python SDK achieves with extra_body + reasoning_effort.

When thinking is on, the API ignores temperature, top_p, presence_penalty, and frequency_penalty. aicommit2 omits temperature and topP from the request in that case so configuration matches API behavior.

Defaults

Model / alias Default thinking
deepseek-v4-flash enabled
deepseek-v4-pro enabled
deepseek-reasoner enabled (legacy alias)
deepseek-r1, deepseek-r1-* enabled
deepseek-chat disabled (legacy alias)
Other names disabled

Override with explicit config:

aicommit2 config set DEEPSEEK.thinking=false
aicommit2 config set DEEPSEEK.reasoningEffort=max

Configuration

DEEPSEEK.key

The DeepSeek API key. If you don't have one, please sign up and subscribe in DeepSeek Platform.

DEEPSEEK.model

Default: deepseek-v4-flash

You can use any DeepSeek model name. The system no longer validates specific model names, allowing you to use new models as soon as they become available.

Current API models and roles (see Models & Pricing):

  • deepseek-v4-flash — Default recommendation; 1M context, up to 384K output tokens; cost-effective.
  • deepseek-v4-pro — Higher capability; pricing and discounts are documented on the pricing page.
  • deepseek-chat — Legacy alias for deepseek-v4-flash in non-thinking mode; DeepSeek has indicated these names may be deprecated later.
  • deepseek-reasoner — Legacy alias for deepseek-v4-flash in thinking mode; same deprecation note.

Tip

We gently recommend using the newer model names (like deepseek-v4-flash) instead of the legacy chat and reasoner names. As noted in the official pricing documentation:

The model names deepseek-chat and deepseek-reasoner will be deprecated in the future. For compatibility, they correspond to the non-thinking mode and thinking mode of deepseek-v4-flash, respectively.

aicommit2 config set DEEPSEEK.model="deepseek-v4-flash"
# or high-capacity model
aicommit2 config set DEEPSEEK.model="deepseek-v4-pro"
# legacy aliases (still accepted by the API for now)
aicommit2 config set DEEPSEEK.model="deepseek-chat"
aicommit2 config set DEEPSEEK.model="deepseek-reasoner"

DEEPSEEK.thinking

Optional boolean. When unset, aicommit2 picks a default from the model name (see table above).

DEEPSEEK.reasoningEffort

Optional: high or max. Used when thinking mode is enabled; defaults to high if unset.