Local AI Model Recommendations en

Local AI Model Recommendations (EN)

Language: Português (Brasil) | English

Goal

Define practical parameter-count and context-window tiers for running FrameCode VibeWork with good performance in a local environment.

Recommended tiers

Tier	Parameters	Context window	Recommended framework usage
Minimum operational	7B-8B	16k	Small fixes, quick edits, and narrow-scope changes
General recommended	14B-32B	32k-64k	Full flow with plan + changelog + wiki, medium sessions, multi-file review
High reliability	32B-70B+	64k-128k	Large refactors, audits, complex releases, and long sessions

Practical rules

• Keep 40% to 60% free context headroom for history, diffs, and long responses.
• For governance-heavy and structured technical writing with less rework, prefer 14B+.
• The minimum tier (16k) is based on the framework's real documentation footprint in multi-document sessions.

Internal framework reference

These ranges were inferred from:

• README - token consumption by scenario
• CONTEXT_MAP.md
• AI.md