Custom model support

Configure which AI model to use for each Nimbus role. Supports Anthropic (default), OpenAI, and Ollama (local).

View current configuration

nimbus models

Output:

  nimbus models

  planner         claude-opus-4-6          (default)
  implementer     claude-sonnet-4-6        (default)
  reviewer        gpt-4o                   (custom)
  chat            claude-sonnet-4-6        (default)

Configuration

In ~/.nimbus/config.toml:

[models]
planner     = "claude-opus-4-6"
implementer = "claude-sonnet-4-6"
reviewer    = "claude-haiku-4-5-20251001"
chat        = "claude-sonnet-4-6"

# OpenAI (requires: pip install openai + OPENAI_API_KEY)
[models.openai]
enabled = false
api_key = ""

# Ollama — local models, fully offline
[models.ollama]
enabled  = false
base_url = "http://localhost:11434"
planner  = "deepseek-coder:33b"

Supported models

Anthropic (default):

• claude-opus-4-6 — most capable, used for planning
• claude-sonnet-4-6 — fast and capable, used for implementation
• claude-haiku-4-5-20251001 — fastest, used for review

OpenAI (requires pip install openai):

• gpt-4o
• gpt-4o-mini
• gpt-4-turbo

Ollama (local, air-gapped):

• Any model available in your Ollama instance
• Examples: deepseek-coder:33b, codellama:34b, llama3:70b

When to use Ollama

Ollama is ideal for:

• Air-gapped environments where no data can leave the machine
• Organizations with strict data residency requirements
• Experimenting with open-source models
• Reducing API costs for high-volume usage

Install Ollama from ollama.ai, pull a model (ollama pull deepseek-coder:33b), then set enabled = true in your config.