OllamaResearcher

OllamaResearcher - Private Local AI Research

Local, private AI research using your self-hosted Ollama server for sensitive information and offline tasks.

Why Use Ollama?

Use Ollama for:

• Sensitive Information: Financial records, medical data, personal information
• Offline Work: No internet required, works on local network
• Privacy: Data never leaves your local network
• No API Costs: Free to use, no quota limits
• Older Knowledge: Tasks that don't require up-to-date information

Don't use Ollama for:

• Current events or recent information (use Perplexity/Claude instead)
• Web research requiring internet access
• Tasks requiring latest model capabilities

---

Configuration

Ollama Server Details:

• URL: http://localhost:11434/ (your local machine)
• Model: gemma4:31b (default), dolphin3:8b-llama3.1-fp16 (fast)
• Location: localhost (your local machine running Ollama)

No API keys required - runs on your local network.

---

Workflow Routing

| Workflow | Trigger | Behavior | |----------|---------|----------| | Ollama Query | "use ollama to...", "ollama research", "ask ollama", "private AI" | Use Ollama for question/research | | Sensitive Data Analysis | Working with financial/medical data | Automatically suggest Ollama | | Offline Work | When internet access limited | Fall back to Ollama |

---

Examples

Example 1: Financial Analysis

User: "Analyze my investment portfolio using ollama"
→ Routes to Ollama server (keeps financial data local)
→ Uses dolphin3:8b-llama3.1-fp16 model
→ Returns analysis without sending data to cloud

Example 2: Medical Information

User: "Help me understand my medical test results using local AI"
→ Automatically uses Ollama for privacy
→ Processes sensitive health data locally
→ Returns explanation without cloud services

Example 3: General Query with Privacy

User: "Use ollama to explain quantum computing"
→ Sends query to local Ollama server
→ Gets response from dolphin3 model
→ Works offline, no external API calls

---

Tools

ollama-query.ts

Query the Ollama server with a prompt.

bun ${PAI_DIR}/skills/OllamaResearcher/tools/ollama-query.ts "your question here"

Features:

• Direct connection to local Ollama server
• Streaming responses for real-time output
• Automatic error handling and retries
• Privacy-focused (no external API calls)

---

Technical Details

Model Information:

• Name: dolphin3:8b-llama3.1-fp16
• Size: 8 billion parameters
• Precision: FP16 (full precision)
• Base: Llama 3.1
• Specialization: General-purpose reasoning and analysis

API Endpoint:

POST http://localhost:11434/api/generate

Capabilities:

• Text generation and completion
• Question answering
• Analysis and reasoning
• Code generation
• Creative writing

Limitations:

• No internet access (knowledge cutoff from training)
• Slower than cloud APIs
• Limited to model's training knowledge
• No real-time information

---

Privacy & Security

✅ Data stays local - Never sent to external services ✅ No logging - Queries not tracked by third parties ✅ HIPAA/PCI friendly - Suitable for sensitive data ✅ Network isolated - Only accessible on local network ✅ No API keys - No credentials to leak

Use Ollama for sensitive data processing without privacy concerns.

---

Performance

Expected Response Times:

• Simple queries: 1-3 seconds
• Complex analysis: 5-15 seconds
• Long-form generation: 15-30 seconds

Depends on:

• Server hardware specs
• Model size and complexity
• Length of input/output
• Network latency (local network)

---

Troubleshooting

Server not responding:

# Check if server is accessible
curl http://localhost:11434/api/tags

# Test model availability
curl http://localhost:11434/api/show -d '{"name":"dolphin3:8b-llama3.1-fp16"}'

Model not found:

• Verify model name is exactly: dolphin3:8b-llama3.1-fp16
• Check available models: curl http://localhost:11434/api/tags

Network issues:

• Ensure you're on the same network as the server
• VPN may block access to local IPs
• Firewall rules may need adjustment

---

This completes the OllamaResearcher skill documentation.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: CarbeneAI
• Source: CarbeneAI/Forge
• License: MIT
• Homepage: http://carbene.ai/open-source

Install and usage instructions live in the source repository linked above.