| name | skill-opencode |
| description | Model server administrator for managing OpenCode providers, GPU optimization, and local LLM endpoints. - opencode - local_models - model_management - install_models - check_hardware - "@mcp-manager" - "@gemini-conductor" |
Opencode Manager (GPU Accelerated)
Expert in managing the opencode platform for local LLM execution. Optimized for NVIDIA RTX 3060 (6GB VRAM).
Core Management
check-hardware
Detects GPU/CPU/RAM and recommends the best SOTA models for the current tier.
- Command:
python3 {PROJECT_DIR}/bin/model_selector.py check
install-best-models
Downloads and installs the GPU-optimized 2026 model stack.
- Command:
python3 {PROJECT_DIR}/bin/model_selector.py install
Platform Commands
- Serve:
opencode serve (Starts headless server)
- Web:
opencode web (Opens browser TUI)
- Stats:
opencode stats (Token/Cost usage)
High-Performance 2026 Stack
| Model | Usage | Target Device |
|---|
| Qwen 2.5 Coder 7B | Primary Agentic Work (Tools/Coding) | GPU (Fits in 6GB) |
| DeepSeek R1 8B | Complex Reasoning / Math / Logic | GPU (Fits in 6GB) |
| Gemma 4 E4B | Ultra-fast Edge Reasoning | GPU (Instant) |
| Llama 3.2 3B | Lightweight Chat Fallback | GPU/CPU |
Performance Note
Local execution on this system is currently configured for GPU Acceleration via CUDA 12.7. Expect speeds of 30-50 tokens/second on primary models.
