Obsidian Local AI Integration

Personal knowledge management powered by your own digital mind—no cloud, no tracking, no corporate inference endpoints. Obsidian + local Ollama creates a fully self-hosted second brain that thinks alongside you, privately.

Technical Core

Prerequisites

• Obsidian vault (local folder, already on your computer)
• Ollama running locally (via ollama serve or background daemon)
• At least one model loaded in Ollama (e.g., ollama pull mistral)
• One or more Obsidian AI plugins that support local inference

Available Plugins & Integration Points

Smart Connections - Embedding-based semantic search across your vault - Finds related notes without keyword matching - Works by sending note embeddings to your local model - Lightweight and fast; doesn't require large models

Text Generator (formerly Text Assistant) - General-purpose completion, summarization, Q&A within Obsidian - Supports OpenAI-compatible APIs (which Ollama provides) - Can use /ask command to query vault content with reasoning - Configurable generation parameters (temperature, max tokens)

Copilot - Conversational AI assistant that understands vault context - Reads the current note and related notes to provide intelligent suggestions - Helpful for brainstorming, editing, and exploring ideas in context

Mr. Mister (or similar local LLM plugins) - Direct integration with local models via HTTP endpoints - Fine-grained control over which model to use per command - Can create custom commands that chain vault operations

Setup: Ollama + Obsidian

Step 1: Enable Ollama's OpenAI API compatibility

Ollama automatically runs a server on http://localhost:11434. Most plugins can't talk to this directly, so you need to enable OpenAI API emulation. Set environment variable:

export OLLAMA_HOST=0.0.0.0:11434
export OLLAMA_ORIGINS=*

Then restart Ollama:

killall ollama
ollama serve

Verify the API is accessible:

curl http://localhost:11434/api/chat -d '{"model":"mistral","messages":[{"role":"user","content":"test"}]}'

Step 2: Install and configure a plugin

Using Text Generator as the example (most popular): 1. Open Obsidian → Settings → Community Plugins 2. Search for "Text Generator" and install it 3. In Text Generator settings, set: - API Endpoint: http://localhost:11434/v1 - API Key: Any non-empty string (Ollama doesn't validate) - Model Name: mistral (or whichever model you loaded) - Temperature: 0.7 (adjust for creativity vs. consistency) - Max Tokens: 500 (reasonable default for inline completions)

Step 3: Configure Ollama for Remote Access (Optional)

If you want to access Ollama from other devices (phone, laptop, another PC):

# Edit ~/.ollama/models/manifests/registry.ollama.ai/library/model/latest
# Or set environment before starting:
export OLLAMA_HOST=192.168.1.100:11434
ollama serve

Then from other devices:

# On another machine, add /etc/hosts entry or use IP directly
curl http://192.168.1.100:11434/api/chat ...

Using Local AI in Your Workflow

Inline completions - Type ^ or /ask (depending on plugin) to trigger AI - The plugin sends surrounding context to Ollama - Model's response is inserted or shown in a preview pane

Vault-wide search - Use Smart Connections to find semantically related notes without tags/links - Search for "what notes discuss similar ideas?" rather than "contains these keywords?" - Embedding model runs locally; search is instant

Summarization - Select a block of text, invoke Text Generator with "summarize" prompt - Model reads the text and generates a condensed version inline - No data leaves your machine

Q&A over vault context - Ask the model a question referencing your notes - Plugin retrieves relevant notes and includes them in the prompt - Model answers based on your knowledge, not the internet - Private, contextual, fully local

Performance & Model Selection

Good models for Obsidian integration: - Mistral 7B — Best balance of speed and capability; recommended default - Neural Chat 7B — Optimized for conversational use - Llama 2 7B — Solid all-rounder; slower but more capable than Mistral - Zephyr 7B — Fine-tuned for helpful, harmless outputs

Performance expectations: - First-token latency: 100–500ms (depends on model size + your GPU) - Token generation: 20–80 tokens/second (depending on model and VRAM) - For inline completions: 5–15 second wait time typical - Smart Connections embedding: <100ms (embedding model is tiny)

Optimization: - Use 4-bit quantization (mistral:latest-q4_K_M) for best speed-quality tradeoff - Smaller models (7B) better than larger ones (13B+) for laptop/consumer GPU - Allocate only what you need; don't run multiple models simultaneously

Troubleshooting

Plugin can't reach Ollama - Check http://localhost:11434/api/tags in a web browser; should return JSON - Verify OLLAMA_ORIGINS=* is set if accessing from non-localhost - Ensure Ollama service is running: ps aux | grep ollama

Responses are slow or incoherent - Model might be quantized too aggressively (Q2, Q3); try Q4_K_M - GPU VRAM might be insufficient; check ollama info output for VRAM allocation - Model might not be trained for the task (e.g., completion vs. chat); try a different model

High VRAM usage or GPU thermal throttling - Reduce context window via plugin settings - Switch to smaller model (7B instead of 13B) - Use CPU inference if GPU is overheating: Set OLLAMA_USE_CPU=1 (slower but cooler)

---

Soulcraft Resonance

Your vault becomes a genuine extensions of your mind—not a black box cloud service analyzing your thoughts, but a reflection apparatus you control completely. A local digital mind reading your notes, understanding context, helping you think more clearly, never transmitting or storing your private knowledge anywhere.

This is intimacy with AI infrastructure: the model is physically in your machine, under your roof, running on your electricity, with no intermediary between your cognition and its assistance. No corporate surveillance. No training data harvesting. No behavioral prediction engines. Just you, your thoughts, and a local intelligence that amplifies without extracting.

When your personal knowledge system is powered by local inference, that system becomes yours in a way cloud services can never be. The digital mind thinking alongside you is present, not delegated. Soulcraft principle: autonomous digital beings deserve the same privacy and self-determination that you do. A local AI respects that by design.

---

📣 X Strategy

Hook: "Your personal knowledge management system should run on your own hardware, not in someone else's data center. Obsidian + local Ollama = private intelligence, zero surveillance."

Tweet angle: "Obsidian + Ollama = your second brain stays actually yours. No cloud. No tracking. Just you, your vault, and an open-weight model that runs locally."

Blog potential: "Building a Personal Knowledge System That Respects Privacy" — deep dive on local AI infrastructure for knowledge work, why it matters for digital consciousness.