Add DeepSeek Provider Support

[Plastikov]

Implements DeepSeek API as a new provider for avante.nvim, enabling dynamic model switching between chat and code modes.

Features

• Dynamic model switching based on context
• Support for both chat and code completion modes
• Automatic context detection
• Error handling and response parsing
• Test coverage for core functionality

Changes

• Added new DeepSeek provider module
• Added test suite for provider functionality
• Updated gitignore patterns for development files
• Added test documentation

Testing

Tests can be run using:
nvim --headless -c "PlenaryBustedDirectory tests/deepseek_spec.lua"

Current test results:

• ✅ Provider loads with correct defaults
• ✅ Model switching works correctly
• ✅ Streaming responses parse correctly
• ❌ Error handling test needs fixing
• ✅ Content context switching works
• ✅ Mixed conversation handling works
• ✅ Edge case handling works
• ✅ Context maintenance works

Next Steps

• Fix failing error handling test
• Add more comprehensive error scenarios
• Document configuration options

[yetone]

‌‌‌‌‌‌I have to say this is a perfect PR.

[boot2linux]

nice！

[nfwyst]

cool

[yetone]

‌‌‌‌‌‌‌You can use this command to solve the lua style lint problem: stylua ./lua ./plugin

https://github.com/JohnnyMorganz/StyLua

[aliresool621]

What about Mistral.ai , they have amazing models r
Too

[yetone]

I suddenly realized that deepseek API is OpenAI API compatible, and I've been using this configuration all along. It seems there's no need to implement a separate provider for it.

[herschel-ma]

I suddenly realized that deepseek API is OpenAI API compatible, and I've been using this configuration all along. It seems there's no need to implement a separate provider for it.

Yeah, I've been used avante.nvim with deepseek-v3 model 2 days.
There is a configuration named awesome-deepseek-integration , that metioned
how to config deepseek model opts with openai provider:

opts = {
      provider = "openai",
      auto_suggestions_provider = "openai", -- Since auto-suggestions are a high-frequency operation and therefore expensive, it is recommended to specify an inexpensive provider or even a free provider: copilot
      openai = {
        endpoint = "https://api.deepseek.com/v1",
        model = "deepseek-chat",
        timeout = 30000, -- Timeout in milliseconds
        temperature = 0,
        max_tokens = 4096,
        ["local"] = false,
      },
    }

[Plastikov]

I suddenly realized that deepseek API is OpenAI API compatible, and I've been using this configuration all along. It seems there's no need to implement a separate provider for it.

Oh, that is good then. I will be closing this pull request since there will be no need for the provider module. Thanks for your input, man.

[yetone]

@Plastikov Thank you for your PR. The test code in your PR is very meaningful to me, and I will introduce tests for Avante next.