Together AI RAG relevance brief
Read-only Together AI RAG preparation workflow. It embeds explicit text inputs, reranks explicit documents for one query, tables relevance scores, and summarizes retrieval readiness. It never fetches remote documents, uploads files, creates batches, starts fine-tuning, mutates deployments, or calls arbitrary Together AI routes.
How it works
Inspect the data fetches, transforms, gates, and output this recipe runs.
Process (9 steps)
embedding
?
rerank
?
results
default
Apply default
embedding_rows
default
Apply default
result_count
count
Count items in results
embedding_count
count
Count items in embedding rows
summarize
?
rerank_table
to_table
Format results as a data table
card
to_summary
Format results as a summary card
Settings
Configurable at install. Defaults shown — change them anytime in Recued.
focus
setting
=
Summarize reranked relevance, likely best source documents, gaps, ambiguity, and whether the document set is ready for RAG grounding.
query
setting
=
top n
setting
=
5
documents
setting
=
max length
setting
=
800
togetherai
setting
=
rerank model
setting
=
Salesforce/Llama-Rank-v1
embedding model
setting
=
togethercomputer/m2-bert-80M-8k-retrieval
Trust & control
What installing this recipe would let it do. Recued grants these permissions at install — review them there before approving.