warning
Legacy managed-runtime reference. Command examples below describe the older desktop/managed llama.cpp preview and are legacy, not the current BCCLI v0.4 command surface. Use BCCLI and Local AI for current instructions.Coding
What is it?
Coding models are explicitly trained or tagged for software tasks: code generation, refactoring, bug fixing, repository-level reasoning. They are normal GGUF chat models with code-heavy training data.
When should I use one?
- Backend for coding assistants (Integrations) — OpenCode, Cline, Continue.
- Terminal-first workflows via CLI chat.
- Scripted automation through the local API.
Prerequisites
- A model tagged for code tasks — filter the catalog by Coding.
- Enough memory: capable coding models are typically 7B+; 14B+ is meaningfully better.
- For agentic tools: a model with tool-calling support — Tool use.
Minimal working example
powershell
botconnector models search coder --capability coding
botconnector get <publisher>/<model>@Q4_K_M
botconnector run <publisher>/<model>Then point your editor tool at the local endpoint:
http://127.0.0.1:11435/v1Config examples per tool: Integrations.
Choosing size vs speed
| Machine | Suggested model size |
|---|---|
| 8 GB RAM | 1.5–7B at Q4 |
| 16 GB RAM | 7–14B at Q4 |
| 24 GB+ RAM / VRAM | 14–30B at Q4 |
A smaller model that responds quickly often beats a large model that thrashes memory. See Context and memory.
Limitations
- Context length decides how much code fits in one request — see the model's context length on its detail page.
- Agentic coding tools need tool-calling; text-only models answer once and cannot call tools.
- Long files exceed context; split or use a larger-context model.
Troubleshooting
Wrong language, broken syntax, forgotten instructions: Model problems.