Ollama Consult

Consult a local LLM (Ollama) for planning, approach comparison, and decision validation. The local LLM acts as a fast, private sounding board—not a source of ground truth.

When to Use

Use consult_local_llm when:

• Brainstorming multiple implementation approaches before choosing one
• Validating a plan against tradeoffs (performance vs simplicity, etc.)
• Checking whether a direction makes sense before diving deep
• Generating alternative perspectives on a design problem
• Lightweight reasoning that doesn't require Claude's full capabilities

Do NOT use for:

• Security-sensitive analysis (credentials, auth flows, vulnerabilities)
• Tasks requiring up-to-date knowledge (latest APIs, recent events)
• Authoritative technical decisions—always validate the output
• Large codebases or complex multi-file reasoning (local LLMs have smaller context)

How to Consult Effectively

1. Ask Specific Questions

Bad:  "Should I use React?"
Good: "I'm building a dashboard that needs real-time updates and has 3 devs
       familiar with Vue. Should I use React or Vue? Key constraints: 6-week
       deadline, no SSR needed."

2. Provide Relevant Context

Use the context parameter to share background information:

question: "Which database indexing strategy fits this access pattern?"
context: "Table has 10M rows. Primary queries: lookup by user_id (80%),
          range scan by created_at (15%), full-text search (5%).
          PostgreSQL 15, < 500ms p99 latency required."

3. Request Structured Output

Ask for comparisons, pros/cons, or numbered options:

"List 3 approaches for [X]. For each: one-line summary, main advantage,
 main disadvantage."

Interpreting Results

The local LLM provides a perspective, not a verdict:

• Cross-check important recommendations with documentation or tests
• Use the output as a starting point for deeper analysis
• The model may be outdated or hallucinate—verify before implementing

Check Setup First

Before consulting, verify Ollama is running and configured:

1. Call list_models to confirm connection and available models
2. Ensure .claude/settings.local.json has env.OLLAMA_MODEL set to your desired model
3. If Ollama is not running: ollama serve

Example Workflow

User: "Redisとin-memoryキャッシュ、どちらが良い？"

1. Call consult_local_llm:
   question: "Redis vs in-memory cache for session storage. Which fits better?"
   context: "Single-instance Node.js app, ~1000 concurrent users, sessions
             expire in 30min, no horizontal scaling planned yet."

2. Review the local LLM's tradeoff analysis

3. Synthesize with your own knowledge and present to user