Memory

Each workspace has a vector-backed long-term memory. Save facts, decisions and notes; retrieve them by meaning (not just keywords). Agents search it automatically via rag_search, and you can read and write it directly over the API.

Save a memory

POST/v1/workspaces/{workspace_id}/memory —

fields: content (required, the text to remember), and optional kind, title, tags (array of strings), source.

bash

curl https://nexevo.ai/v1/workspaces/$NEXEVO_WORKSPACE/memory \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Refunds over 30 days require a manager approval.",
    "kind": "policy",
    "tags": ["refunds", "support"]
  }'

Search

GET/v1/workspaces/{workspace_id}/memory/search —

query params: q (required, the natural-language query) and top_k (how many results). Results are ranked by semantic relevance.

bash

curl -G https://nexevo.ai/v1/workspaces/$NEXEVO_WORKSPACE/memory/search \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  --data-urlencode "q=what is our refund policy" \
  --data-urlencode "top_k=5"

List & forget

GET/v1/workspaces/{workspace_id}/memory —

list stored memories.

DELETE/v1/workspaces/{workspace_id}/memory/{point_id} —

forget a single memory.

DELETE/v1/workspaces/{workspace_id}/memory —

purge all memory in the workspace (irreversible).

Memory vs Projects vs Corpora: Memory is the agent's evolving recall across all tasks; Projects hold standing context for a specific piece of work; named knowledge collections (corpora) feed larger reference material into the same store. Agents pull from memory automatically — saving good facts here makes every future task smarter.