--- title: "Coze Integration" description: "Connect flows to Coze, ByteDance's AI agent and chatbot platform. Agents chat with published bots, manage conversations and messages, run workflows and chatflows, upload files, and maintain retrieval knowledge bases." url: https://flowrunner.ai/integrations/coze date_modified: 2026-08-07T17:20:23-07:00 --- # Coze [AI](https://flowrunner.ai/integrations/category/ai-llms) Connect flows to Coze, ByteDance's AI agent and chatbot platform. Agents chat with published bots, manage conversations and messages, run workflows and chatflows, upload files, and maintain retrieval knowledge bases. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 26 actions · API key · available [Coze website](https://coze.com/) · [Platform Documentation](https://www.coze.com/open/docs) · Capability data verified 2026-08-12 1. A customer message lands in the support channel 2. Agent runs Chat With Bot with the question and the thread's conversation ID for multi-turn context 3. Agent runs Retrieve Chat; the chat is paused in requires\_action on a tool call 4. Agent reads the requested tool call: a change to the customer's account 5. Agent assembles the customer context and the exact change the bot wants to make 6. Support lead approves; agent executes the tool, resumes the chat with Submit Tool Outputs, and posts the bot's answer back to the thread ## What This Integration Enables Coze bots do more than answer: they call tools mid-conversation, and when a tool must run on your side, Coze pauses the chat in a requires\_action state and waits. That pause is the most interesting thing about this connector, because it is a built-in decision point. FlowRunner agents drive conversations with Chat With Bot, maintain persistent history through conversations and messages, run published Coze workflows and chatflows as reusable steps, upload files for chats and knowledge, and keep retrieval datasets current by importing text and web URLs. And at the requires\_action pause, the workflow decides what happens next: execute the tool and resume with Submit Tool Outputs, or stop and put the request in front of a person first. That is the difference between embedding a chatbot and governing one. The bot converses freely; the consequential actions it requests clear a [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) gate before they execute. ### Without FlowRunner **Bot answers stop at information**: Anything requiring an account change gets handed off entirely **Context resets every thread**: Each question starts the conversation from zero **Knowledge goes stale**: The bot cites documents nobody has updated since launch ### With FlowRunner **Bots act, with a gate on consequence**: Routine tool calls execute; consequential ones wait for approval **Multi-turn memory that persists**: Conversations carry full history across chats **Knowledge refreshed by pipeline**: New content lands in datasets as it is published ## Use Case Scenarios ### A support bot that acts, with approval on the line that matters A message arrives in [Slack](https://flowrunner.ai/integrations/slack). The agent passes it to Chat With Bot with the thread's conversation ID so the bot remembers the exchange. Informational answers post straight back. But when Retrieve Chat shows the chat paused in requires\_action because the bot wants to apply an account change, the agent routes the request to the support lead first. On approval it executes the tool, resumes the bot with Submit Tool Outputs, and posts the final answer to the thread. The customer sees one fluid conversation; the account change had a named approver. ### Knowledge that tracks the help center When a new article publishes, the agent runs Create Dataset Document with the page's URL, and Coze parses and chunks it for retrieval. Internal content flows in as text the same way, and Upload File covers documents and images. When articles are retired, the corresponding cleanup runs Delete Dataset Document, gated for review because deletion is permanent and changes what the bot can cite. The bot's answers track what the company actually publishes, not what it published last quarter. ### Coze workflows as audited building blocks A published Coze workflow becomes a callable step: Run Workflow executes it synchronously for quick operations, and Run Workflow Async with Get Workflow Run History handles the long-running ones. After each run, the agent logs inputs, outputs, and status to [Google Sheets](https://flowrunner.ai/integrations/google-sheets) so every automated decision has a row in the audit trail. Run Chatflow does the same for conversational workflows that need dialogue context, carrying a conversation ID across turns. ## Human-in-Loop Highlight Submit Tool Outputs is the action that resumes a paused bot, and what it resumes with is the result of whatever tool the bot asked for. Coze itself holds the chat in requires\_action until those outputs arrive, which means the connector inherits a natural stopping point, and FlowRunner turns it into a real gate. When the requested tool call is consequential, the agent does not execute and resume on its own. It posts the specifics: "The support bot wants to apply a 20 percent goodwill credit to this customer's account, based on the conversation attached. Approve and resume the chat, or take over the thread?" Only after the support lead answers does the tool run and Submit Tool Outputs hand the result back to the bot. The bot stays fluent; the account change gets an owner. 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 26 actions ### Chat 5 - **Chat With Bot** Sends a message to a Coze bot and returns its reply, polling the chat to completion. Supports multi-turn context via a conversation ID; the bot may reason and call tools before answering. - **Retrieve Chat** Retrieves a chat's status: in progress, completed, failed, or requires\_action on a tool call. The check that surfaces the pause point. - **List Chat Messages** Lists all messages generated during a chat, including tool output, reasoning, and follow-up suggestions. - **Cancel Chat** Stops a long-running or runaway chat that has not completed. - **Submit Tool Outputs** Submits client-side tool results back to a chat paused in requires\_action, resuming the bot. Consequential tool calls are routed through human approval in FlowRunner workflows before this runs. ### Conversations 8 - **Create Conversation** Creates a conversation container that stores message history across chats, optionally seeded with initial messages. - **List Conversations** Lists a bot's conversations with pagination. - **List Conversation Messages** Pages through a conversation's stored messages in order. - **Create Conversation Message** Adds a message directly to history without triggering a response, for seeding prior context. - **Get Conversation Message** Retrieves a single message's full details. - **Update Conversation Message** Corrects or augments a stored message. - **Delete Conversation Message** Permanently deletes a message from a conversation. Cannot be undone. - **Clear Conversation** Starts a new context section so later chats no longer see earlier messages, resetting context while keeping the conversation ID. ### Bots 2 - **Get Published Bots** Lists the bots published to the API channel in a workspace, the discovery step before Chat With Bot. - **Get Bot Info** Retrieves a bot's published configuration: prompt, model, plugins, workflows, and knowledge bases. Worth inspecting before trusting a bot with a workflow. ### Files 2 - **Upload File** Uploads a file from FlowRunner to Coze for use in chats or knowledge bases, returning a file ID. - **Get File Info** Retrieves an uploaded file's metadata to verify an upload. ### Workflows 4 - **Run Workflow** Runs a published Coze workflow synchronously with its input parameters and returns the output. - **Run Workflow Async** Starts a long-running workflow and returns an execute ID immediately. - **Get Workflow Run History** Polls an async run for status, output, and errors until it completes. - **Run Chatflow** Runs a conversational workflow with dialogue context, carrying a conversation ID for multi-turn exchanges. ### Knowledge Bases 5 - **List Datasets** Lists the knowledge bases in a workspace with name filtering. - **Create Dataset** Creates a knowledge base to hold documents bots retrieve from. - **List Dataset Documents** Lists a dataset's documents with processing status, the audit surface for knowledge health. - **Create Dataset Document** Adds a document from raw text or a web URL; Coze parses and chunks it automatically. - **Delete Dataset Document** Permanently deletes documents from a dataset so bots can no longer retrieve from them. Cannot be undone. Routed through human approval in FlowRunner workflows. ## Frequently Asked Questions ### What can FlowRunner do with Coze? FlowRunner agents can run Chat With Bot, Retrieve Chat, and List Chat Messages in Coze, plus 23 more actions. ### Does connecting Coze to FlowRunner require OAuth? No. Coze connects to FlowRunner with an API key, no OAuth flow required. ### Can Coze trigger a FlowRunner workflow automatically? Coze doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at Coze?** This integration exposes Coze to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/coze. Site index: https://flowrunner.ai/llms.txt