Continue a run with a new prompt

Node examples use the shared Node client (retriever.mjs). Problem: the run found 30 companies. You now want their emails, in the same context. Solution: start a new run with the first run’s conversation_id.

curl -X POST "$RETRIEVER_API_URL/v1/runs" \
  -H "Authorization: Bearer $RETRIEVER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"conversation_id":"conv_001","prompt":"Add the verified CEO email for each company."}'

import { call, sleep, waitRun } from "./retriever.mjs";

async function followUp(conversationId, prompt) {
  for (;;) {
    try {
      return await call("/v1/runs", { method: "POST", body: { conversation_id: conversationId, prompt } });
    } catch (error) {
      if (error.code !== "conversation_busy") throw error;
      const { items } = await call(`/v1/runs?conversation_id=${conversationId}&limit=1`);
      if (items[0]) await waitRun(items[0].id);
      else await sleep(15_000);
    }
  }
}

The agent keeps the conversation history and tables.
The same applies to a result you want to change: “find 5 more people”, “also include purchasing and CSR roles”, “extend to the subsidiaries”. Continue the conversation instead of starting a new run with a longer prompt, and ask the agent to add to the existing table: it sees what it already found. Leads already in the table (a person by LinkedIn profile, a company by domain) are left out of the searches that add to it and not billed, across sources. Leads of your other tables can come back. A search for the right person at a given company and contacts found row by row in a table are not covered, so check those for duplicates. People who must never come back go in the exclusion list (POST /v1/exclusions/people), not in the prompt. The list is shared by your whole organization and filters what sources bring into every later table. It does not filter Go, direct service calls, contacts found row by row in a table, rows the agent writes itself from a service call, or lists you import into a table.
Only one active run per conversation, otherwise 409 conversation_busy. This also applies if someone is chatting in the app on the same conversation.
Do not send a workspace_id that differs from the conversation’s (400 workspace_mismatch).

Get started

Model Context Protocol

Concepts

Recipes

Continue a run with a new prompt