Dust
AIConnect AI agents to Dust, the enterprise AI agent platform. Agents discover and mention Dust agents, run conversations and poll for replies, keep data source knowledge current by upserting documents and table rows, search it semantically, and execute custom Dust apps.
What This Integration Enables
Dust is where a company builds its AI agents and curates the knowledge they draw on. FlowRunner is the layer that connects those agents to the places work actually arrives. Agents discover and inspect Dust agent configurations, start conversations and poll asynchronous replies, capture quality feedback, and run Dust apps as reusable pipeline steps. On the knowledge side, they keep static data sources current by upserting documents and table rows from other systems, search a data source semantically, and search the whole workspace when they do not know where the answer lives. This is [Orchestration as a Service](/concepts/orchestration-as-a-service/) in its plainest form: FlowRunner does not compete with the agent builder, it coordinates what the builder produces. Dust makes the agents smart. FlowRunner decides when they get invoked, what feeds their knowledge, and which of their outputs a human must see before it counts.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Questions in, governed answers out
A customer question arrives by ticket or email. The agent resolves the right Dust agent with Search Agents, opens a thread with Create Conversation, and polls Get Agent Reply until the answer completes. Instead of firing the reply straight back, it posts the draft with its cited sources to [Slack](/integrations/slack), where the support lead approves, edits, or takes over. Approved answers go out; the agent records the outcome with Submit Message Feedback so answer quality becomes a measured signal, not a feeling.
A knowledge base that stays current on its own
When a new file lands in [Google Drive](/integrations/google-drive-api), the agent runs Upsert Document to index its content into the right static data source, with tags and a source URL for citations. Structured data flows the same way: rows from operational systems land via Upsert Rows into tables agents can query. The knowledge Dust agents answer from stops being a snapshot of the last manual upload and becomes a live reflection of the systems it came from.
Retrieval without a conversation
Not every workflow needs an agent's prose. An agent runs Search Data Source to pull the most relevant chunks for a topic, assembles a cited summary, and sends it to a customer through [Gmail](/integrations/gmail-service). For repeatable multi-step reasoning, Create App Run executes a Dust app built in the Developer Platform and returns per-block outputs, turning a tested prompt pipeline into a callable step inside a larger FlowRunner flow.
Human-in-Loop Highlight
The dangerous operations in this connector are the quiet ones: Upsert Document fully overwrites an existing document with the same ID, Delete Document is permanent, and Delete Table destroys a table and every row in it. These writes change what every Dust agent in the workspace can retrieve, silently and for everyone. So when a sync flow detects documents whose sources have disappeared, the agent does not clean house on its own. It posts the reconciliation to the knowledge owner: "14 documents in the sales-enablement data source no longer exist in the source folder. Delete them? Every agent retrieving from this source loses them permanently. Approve the list, edit it, or keep everything." Deletion runs only on the approved list. Retrieval is cheap to fix; destroyed knowledge is not.
Agent Capabilities
31 actionsAgents
3- List Agents Lists the agent configurations in the workspace with ID, name, description, scope, model, and status. The discovery step before mentioning an agent in a conversation.
- Get Agent Retrieves a single agent's configuration, including instructions, model settings, actions, and status. Useful for inspecting how an agent is set up before relying on it.
- Search Agents Searches agents by name when the exact ID is unknown.
Conversations
9- Create Conversation Starts a new conversation with a Dust agent, seeding it with an initial message that mentions the agent. Supports blocking mode or asynchronous polling, and optional scoping to a space.
- Get Conversation Retrieves a conversation with its ordered user and agent messages. Used to read completed replies or audit a thread.
- Update Conversation Renames a conversation or marks it read.
- Post Message Posts a new user message to an existing conversation, optionally mentioning an agent to generate a reply.
- Edit Message Edits a message's content and mentions. Editing a message that mentions an agent triggers a fresh response.
- Get Agent Reply Polls the event stream for an agent's response to a specific message, resuming from a known event ID to avoid re-reading.
- Get Conversation Events Returns the event stream for a whole conversation, for polling all activity in a thread.
- Cancel Message Generation Stops in-progress agent message generation, the brake for a long-running or unwanted reply.
- Submit Message Feedback Records thumbs-up or thumbs-down feedback on an agent message with an optional comment, turning answer quality into a tracked signal.
Spaces and Data Sources
2- List Spaces Lists the spaces (projects) in the workspace with their IDs and kinds. Spaces contain the data sources, tables, and apps that other operations target.
- List Data Sources Lists the data sources in a space with names, descriptions, and connector details.
Documents and Search
6- List Documents Lists documents in a data source with pagination, for browsing or auditing a knowledge base.
- Get Document Retrieves a single document's full text, tags, and metadata.
- Upsert Document Creates or replaces a document by ID in a static data source, with title, tags, and a source URL for citations. An existing document with the same ID is fully overwritten, so sync flows route destructive overwrites through review.
- Delete Document Permanently deletes a document from a static data source. Routed through human approval in FlowRunner workflows.
- Search Data Source Runs semantic and full-text search over one data source, returning the most relevant chunks with scores. Tag filters narrow the field.
- Search Workspace Searches knowledge nodes across the whole workspace when you do not know which data source holds the answer.
Tables and Rows
8- List Tables Lists the structured tables in a data source with schemas.
- Get Table Retrieves a table's metadata and schema before writing to it.
- Upsert Table Creates or updates a table in a static data source, ready for rows.
- Delete Table Permanently deletes a table and all of its rows. Routed through human approval.
- List Rows Lists a table's rows with pagination.
- Get Row Retrieves one row by ID without paging the whole table.
- Upsert Rows Inserts or updates rows; reusing a row ID overwrites that row. The write path for keeping structured knowledge current.
- Delete Row Deletes a single row by ID.
Apps
3- List Apps Lists the Dust apps available in a space, custom LLM workflows built in the Developer Platform.
- Create App Run Executes a Dust app against a set of inputs and returns per-block outputs. Turns a tested prompt pipeline into a callable workflow step.
- Get App Run Polls an asynchronous app run for status and results.
Frequently Asked Questions
What can FlowRunner do with Dust?
FlowRunner agents can run List Agents, Get Agent, and Search Agents in Dust, plus 28 more actions.
Does connecting Dust to FlowRunner require OAuth?
No. Dust connects to FlowRunner with an API key, no OAuth flow required.
Can Dust trigger a FlowRunner workflow automatically?
Dust doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Dust
$100 in credits. No card required. Connect in minutes.