Configuration & Provider Integrations

Phantom requires an underlying brain. Fortunately, the core architecture is inherently model-agnostic.

We aggressively support the SOTA (State of the Art) across all major frontiers.

The Setup Utility

After executing phantom boot, you config the system by executing:

npx @phantom-pm/cli@latest config setup

This will trigger a terminal UI walking you through injecting your API keys securely into Phantom’s memory banks (~/.phantom/config.json).

Supported Super Intellects

1. OpenAI (The Mainstream Benchmark)

• Supported Models: gpt-4o, gpt-4o-mini, o1, o3-mini
• Setup: Provide your OPENAI_API_KEY.

2. Anthropic (The OS Agent Standard)

• Supported Models: claude-3.7-sonnet, claude-3-opus, claude-3.5-haiku
• Note: Claude 3.7 Sonnet is currently the strictly recommended model if you plan to boot OpenClaws via the Browser Agent due to its native Computer Use xy API blocks.

3. DeepSeek (The Mathematical Vanguard)

• Supported Models: deepseek-chat (V3), deepseek-reasoner (R1)
• Setup: Provide your DEEPSEEK_API_KEY. Perfect for exhaustive PRD logic and rigorous technical requirement generation.

4. Gemini (The Context Behemoth)

• Supported Models: gemini-3.1-pro
• Use Case: Pumping entire Slack histories or 100-page Notion transcripts into Phantom for Rowboat Semantic Mapping.

5. Ollama (The Absolute Privacy Airgap)

• Supported Models: llama3.1:8b, qwen2.5, mistral, deepseek-coder-v2
• Use Case: 100% offline, privacy-centric PMing.
• Setup: Ensure Ollama is running (ollama serve) and phantom will natively intercept localized calls to localhost:11434.

Modifying Configurations Mid-Flight

Phantom allows you to hot-swap intellects on the fly during a single Chat session.

In the CLI Omnibar:

# Force the Chat module to utilize Claude
npx @phantom-pm/cli@latest chat --provider anthropic --model claude-3.7-sonnet