> ## Documentation Index
> Fetch the complete documentation index at: https://docs.jinba.io/llms.txt
> Use this file to discover all available pages before exploring further.
# jinba-flow Integration
> Using workflows as tools, flow runs and approval steps in chat, the Flow run console, and AI operation
Jinba App Neo integrates with jinba-flow in two ways: workspace connections and AI operation. Choose based on what you need (using both is fine).
| Approach | Best for |
| ------------------------------------ | -------------------------------------------------------------------------------------------------------------- |
| **Workspace connection (call-only)** | Calling published workflows as tools for an agent |
| **AI operation** | Letting the AI handle Jinba end to end from chat — "list my flows," "run this one" — with your own permissions |
## Connecting a workspace
Connect a jinba-flow workspace so its published flows can be called as tools for the AI. Connecting takes one click — there are no API keys to issue or paste. Connecting, listing and running all happen on the same page.
"**Flow**" in the sidebar opens the page (its heading reads "**Jinba Flow**"). Its list column has two sections: "**Connected**" and "**Not connected**." The latter lists the jinba-flow workspaces you belong to whose published flows can be reached from outside. Some may not be connectable depending on their visibility settings (see below).
Press "**Connect**" on a workspace row. It moves to the "**Connected**" section, and its published flows become callable as tools. There is no connection-type choice to make: workspace connections are **call-only** — the AI cannot create or edit flows through them.
Add this connection from the Flow section of the agent's configuration panel. The AI can then run the flows as tools.
* You can also connect straight from the **agent detail page**: its Flow section lists your connectable workspaces too, and one click sets up a connection dedicated to that agent ([Attach resources](/en/pages/jinba_app_neo/agents/attach))
* Connecting **from chat** works too: ask the AI ("connect my Jinba Flow workspace") — or let it propose the connection — and a "**Connect Jinba Flow workspaces?**" card appears listing your unconnected workspaces as a checklist. Tick the ones you want and press "**Connect**." In an agent-bound chat, a "**Connection scope**" toggle lets you choose "**Workspace-shared**" or "**This agent only**." Flows connected this way are usable **in the same conversation** — no need to start a new chat. The card lists organization-shared workspaces; workspaces from your personal Jinba account are connected from the Flow page's "**Not connected**" section
* Workspaces set to **Private** on the jinba-flow side cannot be connected: their rows carry a lock mark noting "**Set to Private on the Jinba Flow side**," and trying to connect opens a "**Can't connect**" dialog explaining why. Switching the workspace's visibility to **Unlisted** makes it connectable (this requires an owner — the dialog's "**Open settings in Jinba Flow**" button takes you to the relevant settings screen)
* Connected workspaces and their published flows stay in that same list column. Selecting a workspace name opens its index beside the list: the flow count, "**Last synced**", the connection's status, and the actions "**Refresh**", "**Change display name**" and "**Disconnect**" (plus "**Go to workspace**" for a workspace you belong to with your own jinba-flow account). On a narrow window the index opens as a sheet from the "…" button on the workspace row instead. If a connected workspace can no longer be reached — it was deleted, changed to **Private**, or your access was removed — its row is hidden from the "Connected" list by default and its flows disappear from both the list and flow search until access is restored. To see such rows, open the "…" menu beside the "**Connected**" heading and turn on "**Show inaccessible workspaces**": they appear marked "**Unavailable**", with "Refresh" and "Disconnect" still available
* Connections stay in sync automatically: opening the Flow page or the composer's connector picker checks the jinba-flow side for changes, and new or updated flows are marked with "**NEW**" / "**Updated**" chips. Manual "**Refresh**" in the workspace index is still there when you want to force a refresh
* Which of a connected workspace's flows an agent actually uses can be narrowed **per agent**: in the agent's Flow section, untick "**Use the whole workspace**" and leave only the flows it may call ticked ([Attach resources](/en/pages/jinba_app_neo/agents/attach)). The composer's pickers in that agent's chats follow the selection
* When a flow runs, a [flow input form](/en/pages/jinba_app_neo/chat/interactive-cards) may appear in the chat
## Display names
A connected workspace and an individual flow can each be given a display name that applies **only inside Jinba App** — as the dialog puts it, "Only changes how this connector is displayed in Jinba App. The name in Jinba Flow and syncing are unaffected."
* **A workspace** — "**Change display name**" in its index on the Flow page
* **A flow** — the pencil next to the flow's name in the run console header
While a display name is in use, the jinba-flow name stays visible as "**Original name:** …", and "**Reset to original name**" puts it back.
## Flow runs in chat
When the AI runs a flow, a "**Flow run**" card appears in the chat with the flow's name and live status — "Running…" with the latest progress message and elapsed time, then "**Completed**" or "**Failed**" with the reason (if you have lost access to the Flow workspace, the card says so). Files the flow produces appear as download chips just below the card, with their real filenames, and are collected with the chat's other deliverables.
For runs started through `run_flow` or a flow's MCP tool, the card also has a collapsed "**Inputs (N)**" section below the outputs. Open it to audit the names and values the flow received, including a value the AI supplied rather than one you typed. Up to 24 inputs are shown; a long value is shortened to its first 4,000 characters, with the complete value available in Flow run history. Runs performed through **AI operation** may not include this section.
Files generated earlier in the chat — by a flow or by the AI itself — can be passed as file inputs to follow-up flow runs, so multi-step pipelines work without downloading and re-uploading anything.
## Approval steps (human-in-the-loop)
When a running flow reaches a WAIT\_FOR\_APPROVAL step, the run pauses until someone decides.
* In chat, the flow-run card switches to "**Awaiting approval…**" and shows an approval form: the message the flow author configured, any data to review, a comment box, and one button per configured action (typically **Approve** / **Reject**). An "**Open approval page**" link and the request's expiry are shown as well
* Whether a comment is required is up to the flow author — always, only when rejecting, or not at all. If a required comment is missing, the card says so and you can try again
* The same approval form appears in the Flow run console for runs started from a form there. If you leave the console while a run waits, its "**Run History**" row reads "**Awaiting your response**" — press "**Respond**" to bring the approval card back into the Output pane
* A request that has expired, or was already decided by someone else, is reported on the card
## Flow run console
"**Flow**" in the sidebar is a run console for executing published flows directly from a form, without going through chat.
* Flows are listed under the workspace that publishes them. Open a workspace to see its flows — one workspace at a time — or search across every workspace and flow with the box above the list. Inside a workspace, flows are nested under their jinba-flow folders: each folder row opens and closes with its caret (folders start open, and the ones you close stay closed), shows how many flows it holds including sub-folders, and carries a dot when a closed folder has updates behind it. When two flows in one workspace share a name, the row adds its jinba-flow folder path (or "**Top level**") so you can tell them apart
* Newly published flows carry a "**NEW**" chip and updated ones an "**Updated**" chip. A collapsed workspace that has changes behind it shows an "**Updates**" chip
* Selecting a flow opens its input form; run it and check the results on the spot. If the flow has an approval step, the approval form appears right in the console. A file input that the flow marks as multi-file accepts several files ("**Choose files**" / "**Add files**", up to 10 per field); the same widget appears in the in-chat flow input form
* After a successful run, an AI-written "**Execution Summary**" is generated for the result automatically, and a collapsed "**Details**" zone opens on a "**Report**" tab (a generated report of the result, with "**Download CSV**" / "**Download Excel**" when it holds table or field data) beside "**Data**" and "**Raw Data (JSON)**". "**Generate**" / "**Regenerate**" rebuild them, and an earlier run in the history offers "**Generate summary & report**". Each generation uses your organization's credits and is limited per hour — the console says so when a limit is hit. An organization can have summaries and reports turned off
* "**Open in Jinba Flow**" in the flow's header opens that flow's own page in the Jinba Flow editor, in a new tab. If you are not a member of the flow's workspace on the Jinba Flow side, the link is shown but disabled, with the reason on hover ("**You are not a member of this workspace in Jinba Flow, so it can't be opened**")
* "**Deliverables**" collects the files this flow's past runs produced, newest first, above "**Output**" and "**Run History**". Selecting one opens it in the shared [file preview](/en/pages/jinba_app_neo/chat/agent-activity) — as does a file listed under "**Output Files**" in the output itself, so an Excel or Word deliverable can be read without downloading it
* In "**Run History**", an expanded row shows "**Output**" before "**Input**", and a long input value scrolls inside its own box instead of stretching the row. A run that is still going carries "**Reattach**" (or "**Respond**" when it is waiting on you) to bring it back into the Output pane
* Drag the divider between the list and the console to resize the list; double-click it to go back to the default, or move it with the arrow keys once it has focus. The width is remembered per device
* The console itself is two columns — input and run on the left, output and history on the right — and the divider between them behaves the same way: drag it, move it with the arrow keys or Home / End, or double-click to return to an even split. It starts at half and half, never gives either column less than a quarter of the width or 260 px, and is remembered per device. When the console is too narrow for two columns the output folds into a drawer and the handle is not shown
## AI operation
Turn on "**AI operation**" and the AI can operate Jinba on your behalf — listing, creating, and running flows: anything you can do.
* Turn it on with the toggle on the "**Let the AI create and edit workflows**" card, at the top of the Flow page's list column above the search box. The card is collapsed by default — the chevron opens the description of what the setting covers. A confirmation dialog explains the scope; press "**Allow**"
* It is a **one-time, account-wide setting**: it covers every Jinba Flow workspace you belong to (including ones you join later). There are no per-workspace grants
* If the AI needs it mid-conversation, a "**Let the AI operate your Jinba?**" card appears in the chat — press "**Enable**", confirm the same "**Allow the AI to operate your Jinba?**" dialog with "**Allow**", and the AI picks up right where it left off
* You can turn it off anytime from the same card
- The AI acts with **your own permissions**, and every action is logged
- Credentials are never passed to the AI
- If your organization's settings disable AI code execution, this setting is unavailable — the in-chat card shows "**Unavailable — your organization's settings disable AI code execution.**" instead of the Enable button
## Troubleshooting
* **A connected workspace has disappeared from the list, or shows "Unavailable"** — the Flow workspace may have been deleted, changed to Private, or stopped allowing your access on the jinba-flow side. Such workspaces are hidden from the "Connected" list by default (turn on "**Show inaccessible workspaces**" in the "…" menu beside the heading to see them), and their flows stay hidden while they cannot be reached. If it still exists, ask the owner to switch its visibility to Unlisted, invite you as a member, or add an external access rule as appropriate, then use "Refresh"
* **A "Can't connect" dialog appears when connecting** — the workspace is set to Private on the jinba-flow side, which does not allow external connections. Switch its visibility to Unlisted (requires an owner; ask one if that isn't you)
* **Flow runs never respond** — if no run appears in the jinba-flow run history, the MCP URL may not be reachable from Anthropic's cloud. jinba-flow instances on `localhost` or internal-only hosts cannot be integrated
* **Connected but tools are missing** — check that the flow is **published**. Connections re-sync automatically when you open the Flow page or the composer's connector picker; "Refresh" in the workspace index forces a refresh