[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6CT78AP2QNDVNMF9M153TG2":3,"skill-same:01M6CT78AP2QNDVNMF9M153TG2":103},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":33,"linkOnly":34,"markdownContent":35,"markdownBody":36,"frontMatter":37,"refName":68,"commitSha":69,"skillDirectory":70,"forksCount":71,"repositoryHtmlUrl":72,"repositoryHomepageUrl":6,"ownerHtmlUrl":73,"ownerBlogUrl":74,"ownerDescription":75,"updatedAt":76,"install":77,"resources":90},"01M6CT78AP2QNDVNMF9M153TG2","local-llm",null,"Local LLM inference — Ollama, GGUF models, quantization, hardware matching, igllama.","bkataru\u002Fzeptoclaw","bkataru","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F14833123?v=4","01M6CT4KVBYNZ90GN6AMQKJN65","Baalateja Kataru","01M6CT4KZT0YZ10C1BED1NVP87",2,"MIT","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Flocal_llm\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32],"local llm","inference","http","api","model","ollama","llm inference","command","context","models","default","hardware","qwen3","pattern","2026-11-10 10:12:30",false,"---\nname: local-llm\nversion: 1.0.0\ndescription: Local LLM inference — Ollama, GGUF models, quantization, hardware matching, igllama.\nauthor: Baala Kataru\ncategory: ai\ntriggers:\n  - type: command\n    commands:\n      - \"llm\"\n      - \"ollama\"\n      - \"local-llm\"\n  - type: pattern\n    patterns:\n      - \".*ollama.*\"\n      - \".*local.*llm.*\"\n      - \".*gguf.*\"\n      - \".*quantiz.*\"\nconfig:\n  properties:\n    ollama_host:\n      type: string\n      default: \"http:\u002F\u002Flocalhost:11434\"\n      description: \"Ollama API host\"\n    default_model:\n      type: string\n      default: \"qwen3:4b\"\n      description: \"Default model for inference\"\n    num_threads:\n      type: integer\n      default: 6\n      description: \"Number of CPU threads for inference\"\n    num_ctx:\n      type: integer\n      default: 4096\n      description: \"Context length\"\n  required: []\n---\n\n# Local LLM Inference\n\nRun LLMs locally for privacy, speed, and offline access. Baala has 41+ models in Ollama.\n\n## Hardware Context\n\n**Baala's Setup:**\n- **Desktop (threadripper):** High-end, can run large models\n- **Laptop:** AMD Ryzen 5 5600U, 18GB RAM, integrated Radeon (CPU-only inference)\n\n## Ollama (Primary)\n\n### Basic Commands\n\n```bash\n# List local models\nollama list\n\n# Pull a model\nollama pull qwen3:4b\nollama pull llama3.2:3b\n\n# Run interactive chat\nollama run qwen3:4b\n\n# Run with prompt\nollama run qwen3:4b \"Explain quantum entanglement briefly\"\n\n# Run with system prompt\nollama run qwen3:4b --system \"You are a physics expert\"\n\n# Show model info\nollama show qwen3:4b\n\n# Remove model\nollama rm model-name\n\n# Copy\u002Frename model\nollama cp qwen3:4b my-qwen\n\n# Create custom model\nollama create my-model -f Modelfile\n```\n\n### API Usage\n\n```bash\n# Generate (one-shot)\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fgenerate -d '{\n  \"model\": \"qwen3:4b\",\n  \"prompt\": \"Hello\",\n  \"stream\": false\n}'\n\n# Chat (multi-turn)\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fchat -d '{\n  \"model\": \"qwen3:4b\",\n  \"messages\": [\n    {\"role\": \"user\", \"content\": \"Hello\"}\n  ]\n}'\n\n# Embeddings\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fembeddings -d '{\n  \"model\": \"nomic-embed-text\",\n  \"prompt\": \"Text to embed\"\n}'\n```\n\n### Modelfile (Custom Models)\n\n```dockerfile\nFROM qwen3:4b\n\n# Set parameters\nPARAMETER temperature 0.7\nPARAMETER num_ctx 8192\nPARAMETER stop \"\"\n\n# System prompt\nSYSTEM \"\"\"You are a helpful physics assistant.\"\"\"\n\n# Template (optional)\nTEMPLATE \"\"\"{{ .System }}\n\n{{ .Prompt }}\"\"\"\n```\n\nBuild: `ollama create my-model -f Modelfile`\n\n## Model Selection Guide\n\n### By RAM Available\n\n| RAM | Recommended Models |\n|-----|-------------------|\n| 4GB | qwen3:0.6b, gemma3:270m, granite4:350m |\n| 8GB | qwen3:1.7b, gemma3:1b, phi4-mini, cogito:3b |\n| 16GB | qwen3:4b, llama3.2:3b, deepseek-r1:8b |\n| 32GB | qwen3:8b, llama3.2:7b, mistral:7b |\n| 64GB+ | qwen3-coder:30b, mixtral:8x7b, llama3:70b |\n\n### By Task\n\n| Task | Models |\n|------|--------|\n| Fast chat | qwen3:4b, phi4-mini |\n| Coding | qwen3-coder:30b, deepseek-coder |\n| Reasoning | deepseek-r1:8b, qwen3:4b-thinking |\n| Vision | qwen3-vl:2b, llava, granite3.2-vision |\n| Embeddings | nomic-embed-text, qwen3-embedding |\n| Translation | translategemma |\n| Function calling | functiongemma, gpt-oss |\n\n### Baala's Installed Models\n\n**Embeddings:** qwen3-embedding:8b, qwen3-embedding:0.6b, nomic-embed-text-v2-moe, mxbai-embed-large\n\n**Chat:** qwen3:0.6b\u002F1.7b\u002F4b, gemma3:270m\u002F1b, phi4-mini, cogito:3b, deepseek-r1:8b\n\n**Vision:** qwen3-vl:2b, granite3.2-vision, deepseek-ocr\n\n**Special:** functiongemma (tool calling), translategemma, gpt-oss (13GB)\n\n## GGUF Models (Manual)\n\nFor models not in Ollama registry, use GGUF files directly.\n\n### Download from HuggingFace\n\n```bash\n# Using huggingface-cli\nhuggingface-cli download TheBloke\u002FMistral-7B-v0.1-GGUF mistral-7b-v0.1.Q4_K_M.gguf\n\n# Using hf-hub-zig (Baala's tool)\nhf-hub-zig download TheBloke\u002FMistral-7B-v0.1-GGUF -f mistral-7b-v0.1.Q4_K_M.gguf\n```\n\n### Quantization Levels\n\n| Quant | Size | Quality | Use Case |\n|-------|------|---------|----------|\n| Q2_K | ~25% | Low | Desperate \u002F testing |\n| Q4_K_M | ~45% | Good | **Best balance** |\n| Q5_K_M | ~55% | Better | If RAM allows |\n| Q6_K | ~65% | Great | Near-FP16 |\n| Q8_0 | ~80% | Excellent | Minimal loss |\n| F16 | 100% | Full | Reference |\n\n**Rule of thumb:** Q4_K_M is the sweet spot for most use cases.\n\n### Run with llama.cpp\n\n```bash\n# Interactive chat\nllama-cli -m model.gguf -c 4096 -ngl 0 -i\n\n# Single prompt\nllama-cli -m model.gguf -p \"Hello, world\" -n 100\n\n# With GPU layers (if CUDA)\nllama-cli -m model.gguf -ngl 35\n\n# Server mode\nllama-server -m model.gguf --host 0.0.0.0 --port 8080\n```\n\n## igllama (Baala's Project)\n\nZig-based Ollama alternative. Lighter weight, uses llama.cpp.zig bindings.\n\n```bash\n# Build\ncd igllama\nzig build -Doptimize=ReleaseFast\n\n# Run (API TBD based on current state)\n.\u002Fzig-out\u002Fbin\u002Figllama serve\n```\n\n## RAM Estimation\n\n**Formula:** `RAM ≈ (params × bytes_per_param) + context_overhead`\n\nFor Q4_K_M:\n- 7B model: ~4-5GB\n- 13B model: ~8-10GB\n- 30B model: ~18-22GB\n- 70B model: ~40-45GB\n\n**Context overhead:** ~2MB per 1K context tokens for 7B model.\n\n## Performance Tuning\n\n### CPU-Only (Laptop)\n\n```bash\n# Set thread count (match physical cores)\nOLLAMA_NUM_THREADS=6 ollama run qwen3:4b\n\n# Or in Modelfile\nPARAMETER num_thread 6\n```\n\n### GPU Offloading\n\n```bash\n# Check GPU memory\nnvidia-smi\n\n# Offload layers (more = faster, needs VRAM)\nOLLAMA_GPU_LAYERS=35 ollama run llama3.2:7b\n\n# Full GPU (if fits)\nOLLAMA_GPU_LAYERS=999 ollama run qwen3:4b\n```\n\n### Context Length\n\n```bash\n# Set context length\nOLLAMA_NUM_CTX=8192 ollama run qwen3:4b\n\n# In Modelfile\nPARAMETER num_ctx 8192\n```\n\n**Warning:** Longer context = more RAM. 32K context on 7B model can add 2-3GB.\n\n## Fallback Strategy (Laptop)\n\nWhen cloud APIs fail, use local models:\n\n1. **Primary:** GPT-OSS 20B (MoE, only 3.6B active, ~14GB RAM)\n2. **Backup:** Qwen3-8B (~5GB, 8-12 tok\u002Fs)\n3. **Minimal:** Qwen3-4B (~3GB, 15+ tok\u002Fs)\n\nConfigure OpenClaw fallback:\n```json\n{\n  \"agents\": {\n    \"defaults\": {\n      \"model\": {\n        \"primary\": \"github-copilot\u002Fclaude-opus-4.5\",\n        \"fallback\": \"ollama\u002Fqwen3:4b\"\n      }\n    }\n  }\n}\n```\n\n## Commands\n\n### `llm list`\n\nList all available Ollama models.\n\n**Example:**\n```\nllm list\n```\n\n**Response:**\n```\nAvailable models:\n\nChat:\n  qwen3:0.6b (0.6B)\n  qwen3:1.7b (1.7B)\n  qwen3:4b (4.0B)\n  gemma3:270m (270M)\n  phi4-mini (3.8B)\n\nEmbeddings:\n  qwen3-embedding:8b (8.0B)\n  nomic-embed-text-v2-moe (1.1B)\n\nVision:\n  qwen3-vl:2b (2.0B)\n  granite3.2-vision (3.0B)\n```\n\n### `llm run \u003Cmodel> [prompt]`\n\nRun a model with optional prompt.\n\n**Example:**\n```\nllm run qwen3:4b \"Explain quantum entanglement briefly\"\n```\n\n**Response:**\n```\nQuantum entanglement is a phenomenon where two or more particles become correlated...\n```\n\n### `llm chat \u003Cmodel>`\n\nStart an interactive chat session with a model.\n\n**Example:**\n```\nllm chat qwen3:4b\n```\n\n### `llm pull \u003Cmodel>`\n\nPull a model from Ollama registry.\n\n**Example:**\n```\nllm pull qwen3:4b\n```\n\n### `llm recommend [--ram \u003CGB>] [--task \u003Ctask>]`\n\nRecommend a model based on RAM or task.\n\n**Example:**\n```\nllm recommend --ram 16\n```\n\n**Response:**\n```\nRecommended models for 16GB RAM:\n\n- qwen3:4b (4.0B) - Fast chat, ~3GB RAM\n- llama3.2:3b (3.0B) - Balanced, ~2.5GB RAM\n- deepseek-r1:8b (8.0B) - Reasoning, ~5GB RAM\n\nBest choice: qwen3:4b\n```\n\n### `llm estimate \u003Cmodel> [--ctx \u003Ctokens>]`\n\nEstimate RAM usage for a model.\n\n**Example:**\n```\nllm estimate qwen3:4b --ctx 8192\n```\n\n**Response:**\n```\nRAM estimation for qwen3:4b:\n\nModel size (Q4_K_M): ~2.4GB\nContext overhead (8K tokens): ~16MB\nTotal: ~2.4GB\n\nRecommended: 4GB+ RAM available\n```\n\n## Troubleshooting\n\n**\"CUDA out of memory\"**\n- Reduce GPU layers: `OLLAMA_GPU_LAYERS=20`\n- Use smaller quantization: Q4_K_S instead of Q4_K_M\n- Close other GPU apps\n\n**Slow generation**\n- Check thread count matches CPU cores\n- Use smaller model or lower quant\n- Reduce context length\n\n**Model not found**\n- Check `ollama list`\n- Pull with `ollama pull model:tag`\n- Check model name spelling\n\n**Server won't start**\n- Check port 11434 not in use\n- Check `systemctl status ollama`\n- Check logs: `journalctl -u ollama`\n\n## Configuration\n\n### `ollama_host`\n\nOllama API host. Default: `http:\u002F\u002Flocalhost:11434`\n\n### `default_model`\n\nDefault model for inference. Default: `qwen3:4b`\n\n### `num_threads`\n\nNumber of CPU threads for inference. Default: `6`\n\n### `num_ctx`\n\nContext length. Default: `4096`\n","# Local LLM Inference\n\nRun LLMs locally for privacy, speed, and offline access. Baala has 41+ models in Ollama.\n\n## Hardware Context\n\n**Baala's Setup:**\n- **Desktop (threadripper):** High-end, can run large models\n- **Laptop:** AMD Ryzen 5 5600U, 18GB RAM, integrated Radeon (CPU-only inference)\n\n## Ollama (Primary)\n\n### Basic Commands\n\n```bash\n# List local models\nollama list\n\n# Pull a model\nollama pull qwen3:4b\nollama pull llama3.2:3b\n\n# Run interactive chat\nollama run qwen3:4b\n\n# Run with prompt\nollama run qwen3:4b \"Explain quantum entanglement briefly\"\n\n# Run with system prompt\nollama run qwen3:4b --system \"You are a physics expert\"\n\n# Show model info\nollama show qwen3:4b\n\n# Remove model\nollama rm model-name\n\n# Copy\u002Frename model\nollama cp qwen3:4b my-qwen\n\n# Create custom model\nollama create my-model -f Modelfile\n```\n\n### API Usage\n\n```bash\n# Generate (one-shot)\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fgenerate -d '{\n  \"model\": \"qwen3:4b\",\n  \"prompt\": \"Hello\",\n  \"stream\": false\n}'\n\n# Chat (multi-turn)\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fchat -d '{\n  \"model\": \"qwen3:4b\",\n  \"messages\": [\n    {\"role\": \"user\", \"content\": \"Hello\"}\n  ]\n}'\n\n# Embeddings\ncurl http:\u002F\u002Flocalhost:11434\u002Fapi\u002Fembeddings -d '{\n  \"model\": \"nomic-embed-text\",\n  \"prompt\": \"Text to embed\"\n}'\n```\n\n### Modelfile (Custom Models)\n\n```dockerfile\nFROM qwen3:4b\n\n# Set parameters\nPARAMETER temperature 0.7\nPARAMETER num_ctx 8192\nPARAMETER stop \"\"\n\n# System prompt\nSYSTEM \"\"\"You are a helpful physics assistant.\"\"\"\n\n# Template (optional)\nTEMPLATE \"\"\"{{ .System }}\n\n{{ .Prompt }}\"\"\"\n```\n\nBuild: `ollama create my-model -f Modelfile`\n\n## Model Selection Guide\n\n### By RAM Available\n\n| RAM | Recommended Models |\n|-----|-------------------|\n| 4GB | qwen3:0.6b, gemma3:270m, granite4:350m |\n| 8GB | qwen3:1.7b, gemma3:1b, phi4-mini, cogito:3b |\n| 16GB | qwen3:4b, llama3.2:3b, deepseek-r1:8b |\n| 32GB | qwen3:8b, llama3.2:7b, mistral:7b |\n| 64GB+ | qwen3-coder:30b, mixtral:8x7b, llama3:70b |\n\n### By Task\n\n| Task | Models |\n|------|--------|\n| Fast chat | qwen3:4b, phi4-mini |\n| Coding | qwen3-coder:30b, deepseek-coder |\n| Reasoning | deepseek-r1:8b, qwen3:4b-thinking |\n| Vision | qwen3-vl:2b, llava, granite3.2-vision |\n| Embeddings | nomic-embed-text, qwen3-embedding |\n| Translation | translategemma |\n| Function calling | functiongemma, gpt-oss |\n\n### Baala's Installed Models\n\n**Embeddings:** qwen3-embedding:8b, qwen3-embedding:0.6b, nomic-embed-text-v2-moe, mxbai-embed-large\n\n**Chat:** qwen3:0.6b\u002F1.7b\u002F4b, gemma3:270m\u002F1b, phi4-mini, cogito:3b, deepseek-r1:8b\n\n**Vision:** qwen3-vl:2b, granite3.2-vision, deepseek-ocr\n\n**Special:** functiongemma (tool calling), translategemma, gpt-oss (13GB)\n\n## GGUF Models (Manual)\n\nFor models not in Ollama registry, use GGUF files directly.\n\n### Download from HuggingFace\n\n```bash\n# Using huggingface-cli\nhuggingface-cli download TheBloke\u002FMistral-7B-v0.1-GGUF mistral-7b-v0.1.Q4_K_M.gguf\n\n# Using hf-hub-zig (Baala's tool)\nhf-hub-zig download TheBloke\u002FMistral-7B-v0.1-GGUF -f mistral-7b-v0.1.Q4_K_M.gguf\n```\n\n### Quantization Levels\n\n| Quant | Size | Quality | Use Case |\n|-------|------|---------|----------|\n| Q2_K | ~25% | Low | Desperate \u002F testing |\n| Q4_K_M | ~45% | Good | **Best balance** |\n| Q5_K_M | ~55% | Better | If RAM allows |\n| Q6_K | ~65% | Great | Near-FP16 |\n| Q8_0 | ~80% | Excellent | Minimal loss |\n| F16 | 100% | Full | Reference |\n\n**Rule of thumb:** Q4_K_M is the sweet spot for most use cases.\n\n### Run with llama.cpp\n\n```bash\n# Interactive chat\nllama-cli -m model.gguf -c 4096 -ngl 0 -i\n\n# Single prompt\nllama-cli -m model.gguf -p \"Hello, world\" -n 100\n\n# With GPU layers (if CUDA)\nllama-cli -m model.gguf -ngl 35\n\n# Server mode\nllama-server -m model.gguf --host 0.0.0.0 --port 8080\n```\n\n## igllama (Baala's Project)\n\nZig-based Ollama alternative. Lighter weight, uses llama.cpp.zig bindings.\n\n```bash\n# Build\ncd igllama\nzig build -Doptimize=ReleaseFast\n\n# Run (API TBD based on current state)\n.\u002Fzig-out\u002Fbin\u002Figllama serve\n```\n\n## RAM Estimation\n\n**Formula:** `RAM ≈ (params × bytes_per_param) + context_overhead`\n\nFor Q4_K_M:\n- 7B model: ~4-5GB\n- 13B model: ~8-10GB\n- 30B model: ~18-22GB\n- 70B model: ~40-45GB\n\n**Context overhead:** ~2MB per 1K context tokens for 7B model.\n\n## Performance Tuning\n\n### CPU-Only (Laptop)\n\n```bash\n# Set thread count (match physical cores)\nOLLAMA_NUM_THREADS=6 ollama run qwen3:4b\n\n# Or in Modelfile\nPARAMETER num_thread 6\n```\n\n### GPU Offloading\n\n```bash\n# Check GPU memory\nnvidia-smi\n\n# Offload layers (more = faster, needs VRAM)\nOLLAMA_GPU_LAYERS=35 ollama run llama3.2:7b\n\n# Full GPU (if fits)\nOLLAMA_GPU_LAYERS=999 ollama run qwen3:4b\n```\n\n### Context Length\n\n```bash\n# Set context length\nOLLAMA_NUM_CTX=8192 ollama run qwen3:4b\n\n# In Modelfile\nPARAMETER num_ctx 8192\n```\n\n**Warning:** Longer context = more RAM. 32K context on 7B model can add 2-3GB.\n\n## Fallback Strategy (Laptop)\n\nWhen cloud APIs fail, use local models:\n\n1. **Primary:** GPT-OSS 20B (MoE, only 3.6B active, ~14GB RAM)\n2. **Backup:** Qwen3-8B (~5GB, 8-12 tok\u002Fs)\n3. **Minimal:** Qwen3-4B (~3GB, 15+ tok\u002Fs)\n\nConfigure OpenClaw fallback:\n```json\n{\n  \"agents\": {\n    \"defaults\": {\n      \"model\": {\n        \"primary\": \"github-copilot\u002Fclaude-opus-4.5\",\n        \"fallback\": \"ollama\u002Fqwen3:4b\"\n      }\n    }\n  }\n}\n```\n\n## Commands\n\n### `llm list`\n\nList all available Ollama models.\n\n**Example:**\n```\nllm list\n```\n\n**Response:**\n```\nAvailable models:\n\nChat:\n  qwen3:0.6b (0.6B)\n  qwen3:1.7b (1.7B)\n  qwen3:4b (4.0B)\n  gemma3:270m (270M)\n  phi4-mini (3.8B)\n\nEmbeddings:\n  qwen3-embedding:8b (8.0B)\n  nomic-embed-text-v2-moe (1.1B)\n\nVision:\n  qwen3-vl:2b (2.0B)\n  granite3.2-vision (3.0B)\n```\n\n### `llm run \u003Cmodel> [prompt]`\n\nRun a model with optional prompt.\n\n**Example:**\n```\nllm run qwen3:4b \"Explain quantum entanglement briefly\"\n```\n\n**Response:**\n```\nQuantum entanglement is a phenomenon where two or more particles become correlated...\n```\n\n### `llm chat \u003Cmodel>`\n\nStart an interactive chat session with a model.\n\n**Example:**\n```\nllm chat qwen3:4b\n```\n\n### `llm pull \u003Cmodel>`\n\nPull a model from Ollama registry.\n\n**Example:**\n```\nllm pull qwen3:4b\n```\n\n### `llm recommend [--ram \u003CGB>] [--task \u003Ctask>]`\n\nRecommend a model based on RAM or task.\n\n**Example:**\n```\nllm recommend --ram 16\n```\n\n**Response:**\n```\nRecommended models for 16GB RAM:\n\n- qwen3:4b (4.0B) - Fast chat, ~3GB RAM\n- llama3.2:3b (3.0B) - Balanced, ~2.5GB RAM\n- deepseek-r1:8b (8.0B) - Reasoning, ~5GB RAM\n\nBest choice: qwen3:4b\n```\n\n### `llm estimate \u003Cmodel> [--ctx \u003Ctokens>]`\n\nEstimate RAM usage for a model.\n\n**Example:**\n```\nllm estimate qwen3:4b --ctx 8192\n```\n\n**Response:**\n```\nRAM estimation for qwen3:4b:\n\nModel size (Q4_K_M): ~2.4GB\nContext overhead (8K tokens): ~16MB\nTotal: ~2.4GB\n\nRecommended: 4GB+ RAM available\n```\n\n## Troubleshooting\n\n**\"CUDA out of memory\"**\n- Reduce GPU layers: `OLLAMA_GPU_LAYERS=20`\n- Use smaller quantization: Q4_K_S instead of Q4_K_M\n- Close other GPU apps\n\n**Slow generation**\n- Check thread count matches CPU cores\n- Use smaller model or lower quant\n- Reduce context length\n\n**Model not found**\n- Check `ollama list`\n- Pull with `ollama pull model:tag`\n- Check model name spelling\n\n**Server won't start**\n- Check port 11434 not in use\n- Check `systemctl status ollama`\n- Check logs: `journalctl -u ollama`\n\n## Configuration\n\n### `ollama_host`\n\nOllama API host. Default: `http:\u002F\u002Flocalhost:11434`\n\n### `default_model`\n\nDefault model for inference. Default: `qwen3:4b`\n\n### `num_threads`\n\nNumber of CPU threads for inference. Default: `6`\n\n### `num_ctx`\n\nContext length. Default: `4096`\n",{"name":5,"author":38,"config":39,"version":56,"category":57,"triggers":58,"description":7},"Baala Kataru",{"required":40,"properties":41},[],{"num_ctx":42,"num_threads":46,"ollama_host":49,"default_model":53},{"type":43,"default":44,"description":45},"integer",4096,"Context length",{"type":43,"default":47,"description":48},6,"Number of CPU threads for inference",{"type":50,"default":51,"description":52},"string","http:\u002F\u002Flocalhost:11434","Ollama API host",{"type":50,"default":54,"description":55},"qwen3:4b","Default model for inference","1.0.0","ai",[59,62],{"type":26,"commands":60},[61,24,5],"llm",{"type":32,"patterns":63},[64,65,66,67],".*ollama.*",".*local.*llm.*",".*gguf.*",".*quantiz.*","main","053df4b1eb4317debebb4df0c91bda09fb7464fe","src\u002Fskills\u002Flocal_llm",0,"https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw","https:\u002F\u002Fgithub.com\u002Fbkataru","https:\u002F\u002Fbkataru.bearblog.dev\u002F","⟨e|acc⟩ resident technical wizard | ai\u002Fml\u002Fsys eng | hacking at it till grug says \"software go zoom\" | prev. computational physics in neutrino phenomenology","2026-10-31 07:05:10",{"source":8,"skillName":5,"commands":78,"prompt":88,"zipAvailable":89},[79,82,85],{"tool":80,"command":81},"npx","npx skills add bkataru\u002Fzeptoclaw --skill local-llm",{"tool":83,"command":84},"bunx","bunx skills add bkataru\u002Fzeptoclaw --skill local-llm",{"tool":86,"command":87},"pnpm","pnpm dlx skills add bkataru\u002Fzeptoclaw --skill local-llm","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Flocal_llm\u002FSKILL.md",true,[91,97],{"path":92,"resourceType":93,"resourceTypeDesc":94,"sizeBytes":95,"sourceHtmlUrl":96},"README.md","REFERENCE","参考文档",2101,"https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Flocal_llm\u002FREADME.md",{"path":98,"resourceType":99,"resourceTypeDesc":100,"sizeBytes":101,"sourceHtmlUrl":102},"skill.zig","OTHER","其他",15636,"https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Flocal_llm\u002Fskill.zig",[104,128,142,160,178,190],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":115,"publishedAt":33},"01M6CT88P82Q2WZDWYP1YPVMS8","zig-dev","Zig development workflow — build, test, WASM, benchmarks, release. Baala's primary language.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Fzig_dev\u002FSKILL.md",[110],{"categoryNo":111,"categoryName":112,"taxonomyType":113,"taxonomyTypeDesc":114},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[116,117,118,119,120,121,122,123,124,125,126,127],"zig dev","dev","zig","wasm","workflow","test","development","release","development workflow","benchmarks","build","language",{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":33},"01M6CT862V8PVCBM0WR3WAMSVD","wsl-troubleshooting","WSL2 troubleshooting — DNS fixes, systemd, Windows interop, networking issues.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Fwsl_troubleshooting\u002FSKILL.md",[],[135,136,137,138,139,140,141],"troubleshooting","wsl troubleshooting","windows","networking","issues","dns","systemd",{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":146,"categories":147,"keywords":151,"publishedAt":33},"01M6CT82WP16J4PSQJ63A6HHRE","web-qa","Web app debugging — Chrome headless screenshots, CDN issues, SRI hash fixes, Canvas quirks.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Fweb_qa\u002FSKILL.md",[148],{"categoryNo":149,"categoryName":150,"taxonomyType":113,"taxonomyTypeDesc":114},"01JBSEEDCAT000000000000009","前端开发",[152,153,154,155,139,156,157,158,159],"web qa","qa","web","debugging","chrome","canvas","screenshots","cdn",{"skillNo":161,"name":162,"nameZh":6,"description":163,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":164,"categories":165,"keywords":166,"publishedAt":33},"01M6CT802TTBQ3P3ET96WFWAP4","semantic-search","Local semantic search for memory using Ollama embeddings.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Fsemantic_search\u002FSKILL.md",[],[167,168,32,24,169,26,170,21,171,172,173,174,22,175,176,23,177],"semantic search","search","results","memory","index","vector","embeddings","file","json","text","find",{"skillNo":179,"name":180,"nameZh":6,"description":181,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":182,"categories":183,"keywords":185,"publishedAt":33},"01M6CT7XBBE6NT23CCN0DXVJT4","rust-cargo","Rust\u002FCargo workflow — build, test, publish to crates.io, benchmarks, cross-compile.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Frust_cargo\u002FSKILL.md",[184],{"categoryNo":111,"categoryName":112,"taxonomyType":113,"taxonomyTypeDesc":114},[186,187,188,120,121,189,125,126],"rust cargo","rust","publish","compile",{"skillNo":191,"name":192,"nameZh":6,"description":193,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":194,"categories":195,"keywords":199,"publishedAt":33},"01M6CT7TC2WQN7SY7TB3R1PFWE","planckeon-sites","Deploy planckeon physics sites — Zola static sites, GitHub Pages, LaTeX\u002FKaTeX rendering.","https:\u002F\u002Fgithub.com\u002Fbkataru\u002Fzeptoclaw\u002Fblob\u002Fmain\u002Fsrc\u002Fskills\u002Fplanckeon_sites\u002FSKILL.md",[196],{"categoryNo":197,"categoryName":198,"taxonomyType":113,"taxonomyTypeDesc":114},"01JBSEEDCAT000000000000011","运维与部署",[200,201,202,203,204,205,206,207,208,26,126,32,209],"planckeon sites","github","deploy","deployment","LaTeX","github pages","site","gh","branch","directory"]