FlowRunner
PricingContact
Theme
Start Free

Relevance AI

AI

Orchestrate Relevance AI from FlowRunner. Agents trigger Relevance agents and wait for results, run tools synchronously or async, and manage knowledge sets and documents to keep retrieval data fresh.

15 actions API key available
A new lead lands from a chatbot conversation or web form
Agent packages the lead's context into a message for the qualification agent
Trigger Agent And Wait hands the lead to the Relevance AI agent and collects its reply
Agent checks the reply parses cleanly into a score, a reason, and a draft follow-up
Qualified leads are written to the CRM with the score and reasoning attached
The rep receives the draft follow-up in Slack alongside the lead summary
The rep edits or approves the draft before anything reaches the prospect

What This Integration Enables

Relevance AI is where teams build their AI workforce. FlowRunner is the layer that puts that workforce to work inside governed processes, which is the entire premise of [Orchestration as a Service](/concepts/orchestration-as-a-service/): the platforms building agents multiply, and the coordination layer above them is what makes the collection useful. Through this connector, a FlowRunner flow delegates a task to a Relevance agent and blocks on the answer with Trigger Agent And Wait, or fires asynchronously and collects later with Poll Agent Job. Multi-turn conversations persist across workflow runs by carrying the conversation id, so an agent interaction can span days of a process without losing its thread. The same connector runs Relevance tools as workflow steps and manages the knowledge sets those agents retrieve from: listing documents, inserting fresh records, and retiring sets that no longer reflect reality. Configuration reads through Get Agent and Get Tool mean flows discover the ids and input schemas they need instead of hardcoding them.

Without FlowRunner

Agents on an island Relevance AI agents run well but hand off to nothing; results are copied out by hand
No gate before the customer An agent's draft can reach a prospect without anyone reading it
Stale retrieval data Knowledge sets drift from the source systems they were loaded from

With FlowRunner

Agents inside governed flows Relevance agents are invoked as workflow steps, with their output validated and routed
A human before every send Drafts stop at the rep; approval is the step that releases them
Retrieval data on a sync loop Insert Knowledge Documents refreshes knowledge sets from the systems of record

Use Case Scenarios

Lead Qualification With a Rep at the Gate

New leads arrive from [Chatbase](/integrations/chatbase). For each one, the flow calls Trigger Agent And Wait, and the Relevance qualification agent scores the lead and drafts a personalized follow-up. FlowRunner validates the structure of the reply, writes the score to the CRM, and routes the draft to the owning rep. The rep approves or edits, and only then does the message go out. The agent does the reading and writing; the rep owns the relationship.

Keeping the Agent's Knowledge Current

The support agent's answers are only as good as its knowledge set. On a schedule, the flow gathers current records from the source system, for example documents listed from [Coze](/integrations/coze), and loads them with Insert Knowledge Documents, using stable _id values so updated records overwrite their prior versions instead of accumulating duplicates. List Knowledge Documents runs after the load as verification. The Relevance agent answers from data that matches the systems of record, not from what was true at setup time.

Tool Chains Across Platforms

A data-enrichment step is already built as a Relevance tool, so the flow does not rebuild it. Get Tool reads the input schema, Run Tool executes with the mapped parameters, and the output feeds the next step, which might be a workflow in [Dify](/integrations/dify) or a native FlowRunner branch. Long-running tools switch to Run Tool Async with Poll Agent Job so the flow keeps working while the tool grinds.

Human-in-Loop Highlight

Delete Knowledge Set permanently deletes an entire knowledge set and every document in it, and it cannot be undone. Any Relevance agent that retrieves from that set loses its grounding at the moment of deletion. So when the refresh flow needs to rebuild a set from scratch, it stages the replacement first and then stops: "The pricing knowledge set is being rebuilt. The old set has 3,200 documents; the staged replacement has 3,150 and passes verification. Deleting the old set is permanent, and the sales agent retrieves from it live. Approve the cutover, or hold?" The operations owner approves, and only then does the agent run Delete Knowledge Set on the old set. That [human-in-the-loop](/concepts/human-in-the-loop/) gate exists because deleting the wrong set silently degrades every conversation the dependent agents have afterward.

Agent processes routinely
Detects exception requiring judgment
Clear match Continues automatically
Ambiguous Routes to human via preferred channel
Human decides
Agent resumes with decision

Agent Capabilities

15 actions

Agents

7
  • Trigger Agent Sends a message to a Relevance AI agent to start or continue a conversation, returning conversation and job identifiers immediately. The asynchronous half of the delegation pattern.
  • Trigger Agent And Wait Sends a message and polls until the agent finishes, returning the reply in a single step. The one-call form for flows that need the answer inline; the poll gives up after roughly two minutes and reports the last known status.
  • Poll Agent Job Checks a running agent or tool job and returns accumulated updates. Called repeatedly with the studio and job ids until completion.
  • Get Agent Retrieves an agent's full configuration: name, system prompt, model, and enabled tools. Read before triggering to know what you are delegating to.
  • List Agents Lists the agents in the Relevance AI project with pagination. How flows discover agent ids.
  • List Agent Tasks Lists an agent's conversations with status and latest activity. The returned conversation ids let flows resume a thread days later.
  • Get Agent Task Retrieves a conversation's full message history, newest first: prompts, agent replies, and tool activity. The audit view of what the agent actually did.

Tools

4
  • Run Tool Runs a Relevance tool with input parameters matching its declared schema and waits for the final output. The workhorse for reusing built tools as workflow steps.
  • Run Tool Async Fires a tool and returns a job id immediately, for long-running tools polled later with Poll Agent Job.
  • List Tools Lists the project's tools with pagination, for discovering what already exists before building anew.
  • Get Tool Retrieves a tool's configuration including its input and output schema, so flows call it with exactly the parameters it expects.

Knowledge Sets

4
  • List Knowledge Sets Lists the project's knowledge sets, the datasets agents and tools retrieve from.
  • List Knowledge Documents Lists the documents in a knowledge set with pagination. Used to inspect and verify before and after loads.
  • Insert Knowledge Documents Inserts documents into a knowledge set, creating it if needed. Stable _id values make loads idempotent, which is what keeps refresh loops from breeding duplicates.
  • Delete Knowledge Set Permanently deletes a knowledge set and all its documents. Cannot be undone, and dependent agents lose retrieval immediately, so this call sits behind a human gate.

Frequently Asked Questions

What can FlowRunner do with Relevance AI?

FlowRunner agents can run Trigger Agent, Trigger Agent And Wait, and Poll Agent Job in Relevance AI, plus 12 more actions.

Does connecting Relevance AI to FlowRunner require OAuth?

No. Relevance AI connects to FlowRunner with an API key, no OAuth flow required.

Can Relevance AI trigger a FlowRunner workflow automatically?

Relevance AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Relevance AI

$100 in credits. No card required. Connect in minutes.