docs: explain local model context sizing

[ousamabenyounes]

Summary

• document practical local-model context window sizes for quick, standard, and deep scans
• add Ollama Modelfile, LM Studio, and llama.cpp context configuration guidance
• list common context clipping symptoms and the Strix context knobs users can tune

Validation

• git diff --check -> passed
• docs-only change; no runtime tests run

Closes #286

[greptile-apps]

Greptile Summary

This documentation update explains local-model context sizing and management.

• Adds practical context-window recommendations for quick, standard, and deep scans.
• Documents Ollama, LM Studio, and llama.cpp context configuration.
• Lists clipping symptoms and the Strix environment variables used to tune compaction and tool-output retention.

Confidence Score: 3/5

The documentation should be corrected before merging because its custom 64k Ollama example can cause Strix to budget against a 200k fallback and compact too late.

The newly recommended custom model name may not resolve through model metadata, while the documented configuration leaves Strix's 200k fallback unchanged despite configuring Ollama for only 65,536 tokens.

Files Needing Attention: docs/llm-providers/local.mdx

Important Files Changed

Filename	Overview
docs/advanced/configuration.mdx	Documents the implemented context-management environment variables and their matching defaults.
docs/llm-providers/local.mdx	Adds useful local-context guidance, but the custom Ollama model example can leave Strix using a 200k fallback for a 64k runtime window.

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
docs/llm-providers/local.mdx:61-65
**Fallback exceeds Ollama context**

When the custom `qwen3-vl-64k` name is not present in LiteLLM's model metadata, Strix budgets against its 200,000-token fallback rather than Ollama's 65,536-token window, delaying compaction until requests exceed the server limit and produce context-overflow or lost-state behavior.

```suggestion
```bash
ollama create qwen3-vl-64k -f Modelfile
export STRIX_LLM="ollama/qwen3-vl-64k"
export LLM_API_BASE="http://localhost:11434"
export STRIX_CONTEXT_FALLBACK_TOKENS="65536"
```
```

_{Reviews (1): Last reviewed commit: "docs: explain local model context sizing" | Re-trigger Greptile}

[ousamabenyounes]

Fixed Greptile feedback in b2e7441.

Change:

• Added STRIX_CONTEXT_FALLBACK_TOKENS="65536" to the Ollama 64k example.
• Added guidance to keep Strix fallback context aligned with custom local model server context.

Validation:

• git diff --check -> passed
• docs-only change; no runtime tests run