Sign InGet Started

Local & Offline Models

Taskmaster can run entirely offline using local AI models. This is useful for air-gapped environments, cost-sensitive workflows, or when you prefer to keep data on your machine.

Ollama

Ollama runs open-source models locally with a simple CLI.

Setup:

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3

# Configure Taskmaster to use it
tm models --set-main ollama/llama3

No API key is needed for local Ollama. If you're connecting to a remote Ollama server, set OLLAMA_API_KEY in your environment.

LM Studio

LM Studio provides a desktop app for downloading and running models locally with zero configuration.

Setup:

  1. Download and install LM Studio
  2. Browse and download a model from the built-in model hub
  3. Start the local server in LM Studio
  4. Configure Taskmaster:
tm models --set-main lmstudio/your-model-name

Claude Code CLI

If you have Claude Code installed, Taskmaster can use it directly — no separate API key needed.

tm models --set-main claude-code/sonnet

Claude Code respects your existing configuration (~/.claude/, .claude/, CLAUDE.md) and provides codebase-aware results by analyzing your project structure.

Gemini CLI

Google's Gemini CLI works with a free Google account.

tm models --set-main gemini-cli/gemini-2.0-flash

Gemini CLI provides structured output support and codebase analysis. A Google Cloud Application (GCA) subscription removes rate limits.

OpenAI-Compatible Endpoints

Connect any service that implements the OpenAI API format:

# Set custom endpoint
tm models --set-main openai-compatible/your-model

# Configure the base URL in .env
OPENAI_COMPATIBLE_BASE_URL="http://localhost:8080/v1"

This works with vLLM, text-generation-webui, LocalAI, and other OpenAI-compatible servers.

Choosing a Local Model

Use case Recommended
Best quality, no internet Ollama with a large model
Easy setup, desktop app LM Studio
Already use Claude Code Claude Code CLI
Free, Google account Gemini CLI
Custom infrastructure OpenAI-compatible endpoint

Limitations

Local models may produce lower quality results compared to cloud providers, especially for complex task generation and expansion. Consider using a cloud provider for the main model and a local model as the fallback:

tm models --set-main claude-sonnet-4-20250514
tm models --set-fallback ollama/llama3