Skip to main content

Using Featherless AI With Roo Code

Featherless AI provides access to high-performance open-source models including DeepSeek, Qwen, and other large language models. All models are currently free to use, making it an excellent choice for budget-conscious developers.

Website: https://featherless.ai

Getting an API Key​

  1. Sign Up/Sign In: Go to Featherless AI. Create an account or sign in.
  2. Navigate to API Keys: Access the API keys page in your account.
  3. Create a Key: Generate a new API key. Give it a descriptive name (e.g., "Roo Code").
  4. Copy the Key: Important: Copy the API key immediately. It will only be shown once. Store it securely.

Supported Models​

Roo Code supports the following Featherless models:

  • deepseek-ai/DeepSeek-R1-0528 (Default) - DeepSeek R1 reasoning model with <think> tag support
  • deepseek-ai/DeepSeek-V3-0324 - DeepSeek V3 model
  • moonshotai/Kimi-K2-Instruct - Kimi K2 instruction-following model
  • openai/gpt-oss-120b - GPT-OSS 120B parameter model
  • Qwen/Qwen3-Coder-480B-A35B-Instruct - Qwen3 specialized coding model

Model Capabilities​

All models support:

  • Context Window: ~32,678 tokens
  • Max Output: 4,096 tokens
  • Pricing: Free (no cost for input/output tokens)
info

DeepSeek R1 Models: The DeepSeek R1 models (like DeepSeek-R1-0528) include special reasoning capabilities with <think> tag support for step-by-step problem solving. These models automatically separate reasoning from regular output.

Configuration in Roo Code​

  1. Open Roo Code Settings: Click the gear icon () in the Roo Code panel.
  2. Select Provider: Choose "Featherless AI" from the "API Provider" dropdown.
  3. Enter API Key: Paste your Featherless API key into the "Featherless API Key" field.
  4. Select Model: Choose your desired model from the "Model" dropdown.

Tips and Notes​

  • Free Tier: All models are currently free with no usage costs, making Featherless ideal for experimentation and development.
  • Model Selection: Choose models based on your needs:
    • DeepSeek R1 models: Best for complex reasoning and problem-solving tasks
    • DeepSeek V3: General-purpose model for various tasks
    • Qwen3 Coder: Optimized for code generation and programming tasks
    • Kimi K2: Balanced instruction-following model
    • GPT-OSS: Large general-purpose model
  • OpenAI Compatibility: Featherless uses an OpenAI-compatible API format for easy integration.
  • Limitations: No image support or prompt caching available on any model.