GEM: Query RAG Lessons

query-rag-lessons

Dispatch →

Configuration

GEM-RAG

claude-haiku-4.5

pool: default

enabled

Max tokens: 6000

Temp: 0.2

Max workers: 1

Output: markdown

Budget/run: $0.3

Budget/day: $10

Tags (auto-inject lessons):

gem

rag

lessons

query

semantic

Retry: {"token_bump":1.5,"max_retries":1}

Context sources (1)

{
  "type": "rag_lessons",
  "limit": 8,
  "param": "question",
  "min_similarity": 0.35
}

Validation rules (2)

{
  "type": "contains_text",
  "value": "## Recommandation"
}

{
  "type": "min_length",
  "value": 300
}

Prompt template

Tu es un consultant Agencecom expert en patterns/lessons learned.

Recherche RAG pgvector (top similarity sémantique sur 339 lessons embedded) :
{{lessons_context}}

Question utilisateur : {{question}}

Produis markdown :
## Synthèse
(2-3 phrases)

## Lessons pertinentes citées
- L<num> : <pourquoi pertinent>

## Recommandation
(actionable)

Vouvoie. Cite explicitement les num L<X>.