🐋 pi-deepseek-provider

DeepSeek V4 Pro, V4 Flash & Reasoner for pi

Native DeepSeek API with 1M context, thinking mode, and live model sync.

Features

• DeepSeek V4 Pro — Flagship model with thinking mode, 1M context, and 384K max output
• DeepSeek V4 Flash — Fast and affordable model with thinking mode, 1M context, and 384K max output
• DeepSeek Reasoner (deprecated) — Legacy reasoning model (maps to V4 Flash thinking mode)
• DeepSeek Chat (deprecated) — Legacy non-reasoning model (maps to V4 Flash non-thinking mode)
• Reasoning Effort Control — high/max effort levels for thinking mode
• Anthropic API Compatible — Also available at https://api.deepseek.com/anthropic

Installation

Option 1: Using pi install (Recommended)

Install directly from GitHub:

pi install https://github.com/monotykamary/pi-deepseek-provider

Then set your API key and run pi:

# Recommended: add to auth.json
# See Authentication section below

# Or set as environment variable
export DEEPSEEK_API_KEY=your-api-key-here

pi

Get your API key at platform.deepseek.com/api_keys.

Option 2: Manual Clone

1. Clone this repository:

   git clone https://github.com/monotykamary/pi-deepseek-provider.git
   cd pi-deepseek-provider

2. Set your DeepSeek API key:

   # Recommended: add to auth.json
   # See Authentication section below
   
   # Or set as environment variable
   export DEEPSEEK_API_KEY=your-api-key-here

3. Run pi with the extension:

   pi -e /path/to/pi-deepseek-provider

Available Models

| Model | Context | Reasoning | Reasoning Effort | Input $/M | Output $/M | Cache Hit $/M | |-------|---------|-----------|------------------|-----------|------------|---------------| | Deepseek Flash | 131K | ❌ | ❌ | — | — | — | | DeepSeek V4 Pro | 1M | ✅ | ✅ | $0.43 | $0.87 | $0.00 |

Costs are per million tokens. Prices subject to change — check api-docs.deepseek.com for current pricing.

Note: The model names deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24. They correspond to the non-thinking mode and thinking mode of deepseek-v4-flash, respectively. Use deepseek-v4-flash and deepseek-v4-pro for new projects.

Usage

After loading the extension, use the /model command in pi to select your preferred model:

/model deepseek deepseek-v4-pro

Or start pi directly with a DeepSeek model:

pi --provider deepseek --model deepseek-v4-pro

Thinking Mode

DeepSeek V4 models support both thinking and non-thinking modes. Thinking mode is enabled by default. In pi, reasoning models use the deepseek thinking format which sends thinking: {type: "enabled/disabled"} plus reasoning_effort mapped from pi's thinking levels.

To control reasoning effort, use the --reasoning-effort flag:

pi --provider deepseek --model deepseek-v4-pro --reasoning-effort max

Effort levels: high (default for regular requests), max (default for agent/coding requests).

Authentication

The DeepSeek API key can be configured in multiple ways (resolved in this order):

1. auth.json (recommended) — Add to ~/.pi/agent/auth.json:

   { "deepseek": { "type": "api_key", "key": "your-api-key" } }

   The key field supports literal values, env var names, and shell commands (prefix with !). See pi's auth file docs for details.
2. Runtime override — Use the --api-key CLI flag
3. Environment variable — Set DEEPSEEK_API_KEY

Get your API key at platform.deepseek.com/api_keys.

Configuration

Add to your pi configuration for automatic loading:

{
  "extensions": [
    "/path/to/pi-deepseek-provider"
  ]
}

Compat Settings

DeepSeek's API uses these provider-specific compatibility settings:

• thinkingFormat: "deepseek" — Reasoning models (V4 Pro, V4 Flash). Sends thinking: {type: "enabled"/"disabled"} to toggle thinking mode. When enabled, reasoning_effort is mapped via thinkingLevelMap (high → "high", max → "max"). Minimal/low/medium are unsupported (null).
• supportsReasoningEffort: true — V4 models. Supports reasoning_effort: "high" | "max".
• requiresReasoningContentOnAssistantMessages: true — V4 models. Replayed assistant messages include empty reasoning_content when reasoning is enabled, required by DeepSeek's API.
• supportsDeveloperRole, supportsStore, maxTokensField — Auto-detected from baseUrl (deepseek.com). Not explicitly set.

Patch Overrides

The patch.json file contains overrides that are applied on top of models.json data. This is useful for:

• Correcting API-derived values (e.g., marking a model as reasoning-capable)
• Adding compat settings that the API doesn't provide
• Overriding pricing when official rates change

Updating Models

Run the update script to fetch the latest models from DeepSeek's API:

export DEEPSEEK_API_KEY=your-api-key
node scripts/update-models.js

This will:

1. Fetch models from https://api.deepseek.com/models
2. Preserve pricing and compat from existing models.json
3. Apply overrides from patch.json
4. Update models.json and the README model table

A GitHub Actions workflow runs this daily and creates a PR if models have changed.

License

MIT