Official scripts
Explore our curated collection of scripts
List Dust agents
List the agents available in the signed-in Dust workspace. Returns workspaceId and agents (sId, name, description, scope, status, model provider/id, userFavorite, canEdit, tags). sId is what you pass as `agent` to ask or send_message. Scope tells you where an agent comes from: "global" is a Dust built-in such as dust, helper, deep-dive or analyst, "workspace"/"published" is one your workspace created, "hidden"/"private" is personal. Requires being signed in to dust.tt.
Search Dust conversations
Search the signed-in Dust workspace's conversations by title text. Returns workspaceId and the matching conversations (sId, title, created, updated, spaceName), newest-updated first, with hasMore and nextCursor. sId is the join key for get_messages. The search covers conversation titles, not message bodies, so a word spoken inside a chat will not match unless it is also in the title — and Dust titles conversations automatically, so titles are paraphrases of the opening message. An empty match list is a real answer about the query, not a failure. Requires being signed in to dust.tt.
Delete a Dust conversation
Delete a Dust conversation by its sId — the inverse of ask, for cleaning up conversations a script created. Returns the title it had, deleted, and verified: the conversation is re-read after the delete, so verified:true means it is genuinely gone rather than merely that the request returned 200. A conversation that was already absent comes back as alreadyAbsent:true with deleted:false, so a no-op is never reported as a successful delete. This is destructive and Dust offers no undo — the conversation and its whole message history go.
Ask Dust
Start a new Dust conversation, send one question to an agent, and wait for the answer to finish (up to 3 minutes). Returns the answer text, the conversation and message ids, the conversation URL, which agent answered, and whether generation completed. Requires a signed-in dust.tt session. The agent defaults to "dust", the workspace's general-purpose agent; pass agent to address another one by its sId (list them with list_agents). Dust answers asynchronously, so this polls the conversation until the agent's turn stops changing: completed=false means the 3-minute budget ran out while the agent was still working, and the partial answer is returned rather than an error — re-read it later with get_conversation_messages using the conversationId. This creates a real conversation in the workspace; delete it with delete_conversation if it was only a test.
List Dust conversations
List the signed-in Dust workspace's conversations, newest first. Returns workspaceId and conversations (sId, title, created, updated, unread, hasError, actionRequired, isRunningAgentLoop, spaceId), plus hasMore and nextCursor for paging. sId is the join key for get_conversation_messages. Pass workspaceId to target a specific workspace when the account belongs to several; it defaults to the first workspace on the account. Paginate with cursor (pass back the nextCursor you were given) rather than an offset — the list is ordered by last update, so an offset would skip or repeat rows as conversations move. Requires being signed in to dust.tt.
Get Dust conversation messages
Read a Dust conversation's messages by conversation sId (from list_conversations or search_conversations), oldest first. Returns workspaceId, conversationId, title, a truncated flag, and messages: each with sId, role (user or agent), content, created, rank, plus author (the human's name/email on a user turn) and agent (the mentioned agent's id on an agent turn). The endpoint pages newest-first from the end of the conversation; this script walks that cursor backwards until it has the whole history or hits `limit`, then returns the messages in reading order. truncated is true when limit stopped it before the first message, so the oldest turns are missing. An unknown conversation id fails loudly rather than returning an empty transcript. Requires being signed in to dust.tt.
Send a Dust message
Send a follow-up message into an existing Dust conversation and wait for the agent's reply (up to 3 minutes). Takes the conversationId from list_conversations, search_conversations or ask. Returns the reply text, the ids of both the sent message and the reply, the agent that answered, and whether generation completed. The agent defaults to "dust"; pass agent to address another one (list_agents). Like ask, this polls rather than streaming, and completed=false means the wait ran out while the agent was still working — the partial reply is returned and the rest can be read later with get_conversation_messages. Only the new reply is returned, not the whole thread: use get_conversation_messages for that.