Memory & RAG

Both memory and knowledge bases are organized by namespace — a name you choose: one per customer, per conversation, per agent, per handbook. In a workflow, build it from data, e.g. support-{{$json.body.customerId}}, so each customer gets their own memory.

memory.store saves a piece of text — a message, a fact, a decision — and memory.recall finds the stored items closest in meaning to a query, most relevant first. Items can expire (ttlMs), and once a namespace grows past 30 items it is automatically compacted into a summary.

curl -X POST https://api.directos.dev/api/memory/customer-42/remember \
  -H "authorization: Bearer $DIRECTOS_KEY" \
  -H 'content-type: application/json' \
  -d '{"role": "user", "content": "Prefers invoices in EUR, billed quarterly."}'

curl "https://api.directos.dev/api/memory/customer-42/recall?q=billing%20currency&limit=5" \
  -H "authorization: Bearer $DIRECTOS_KEY"

rag.ingest turns a document into a searchable knowledge base: it extracts the text, splits it into overlapping chunks and indexes each one. rag.query searches it two ways at once — by meaning and by exact keywords — and merges the rankings, so both paraphrased questions and exact product codes find the right passage. Every result names its source document, ready for citations.

# content is the file's bytes, base64-encoded
curl -X POST https://api.directos.dev/api/rag/handbook/ingest \
  -H "authorization: Bearer $DIRECTOS_KEY" \
  -H 'content-type: application/json' \
  -d "{\"name\": \"refunds.pdf\", \"mimeType\": \"application/pdf\", \"content\": \"$(base64 < refunds.pdf)\"}"

curl "https://api.directos.dev/api/rag/handbook/query?q=refund%20window&limit=5" \
  -H "authorization: Bearer $DIRECTOS_KEY"