⚙️ Parametry recallu

Controls data selection for agent context.

Token budget na dotaz 🔒 2000 💡 +tokens = +quality, +cost

Min. relevance pro recall 🔒 0.3 💡 Lower = more coverage, more tokens

Max records per query 10 💡 +records = +detail, +tokens

Temporal decay factor 0.5 💡 Higher = prefers newer data

Half-life (days) How many days = half relevance score

Dedup podobnost 0.85 Merge results above this threshold

Recall tier 🔒 💡 Higher tier = more detail, more tokens

Sensitivity threshold Records above this level require authorization

Compress above (tokens) Summarize records larger than this

Summary preference 0.8 Weight for summaries vs. raw data

Max. % kontextu modelu 30% 💡 More context = less room for response

📊 Estimated Impact

Kvalita odpovědi

80%

Token consumption

40%

Ochrana dat

90%

Restrictions for AI agents — agents cannot change these.