In-Chat Forms and Suggestions

As the AI works, it may show input forms or suggestion cards in the chat. This page explains what each one means and how to respond. Cards that propose keeping something for later — a skill, knowledge, a connector, or a new agent — open with a plain-language line describing what approving will do; the corresponding product term appears alongside as a label (for example “It will be saved as a Skill”). While the AI is waiting on a card that has scrolled out of view, a “Go to the waiting card” pill appears at the bottom of the conversation. Click it to scroll to the card, which is highlighted briefly so you can spot it.

​

Flow input forms

When running a jinba-flow flow, the required inputs are shown as a form.

• Fill in the required fields (marked with *) and press “Run” to execute the flow
• Fields with an AI guess badge contain initial values the AI inferred from the conversation. You can use them as-is or edit them
• File fields take an upload — choose or drop the file the flow needs, or pass a file generated earlier in the chat. A field the flow marks as multi-file accepts several files (“Choose files” / “Add files”, up to 10); “Run” waits until the uploads finish
• Press “Cancel” to abandon the input

​

Flow approval card

If a running flow reaches a step that waits for a person — an approval step (WAIT_FOR_APPROVAL) or a step that asks you a question — the flow-run card (the card that tracks a flow’s run status in the chat) switches to “Awaiting approval…” and shows a form. Requests you can settle with a choice plus a comment (optional or required) are answered right on the card; if the step needs a form that chat cannot display, the card asks you to respond on the Jinba Flow side instead. See jinba-flow integration for how approvals work (comments, actions, and expiry).

​

Multiple-choice questions

When the AI is unsure how to proceed, it may show a question card with options.

• Choose one option per question (or several, when “Multi-select” is shown)
• If no option fits, pick “Other” and type a free-form answer (typing selects it automatically)
• When the AI asks several questions, the card shows them one at a time, with your position (“2 of 3”) in its header. “Next” moves on, “Skip” leaves a question for later, and the arrows in the header (“Previous question” / “Next question”) take you back to change an answer. “Submit” appears on the last question and sends every answer together; it stays disabled, with the note “Some questions are still unanswered,” until each question has an answer
• “None of these fit” (the ✕ in the card’s header) dismisses the question entirely (every question on the card, when there are several) — the AI won’t re-present the same options and follows up a different way instead
• Once you answer, the AI resumes work and the card becomes a single line, “Your answers were received.” It stays open showing each question with the answer you gave until the AI’s next reply arrives, then folds; click the line to read your answers again

​

Jinba AI operation card

When the conversation calls for jinba-flow operations and AI operation is off, a card may appear asking “Let the AI operate your Jinba?”

• Choose “Enable” and an “Allow the AI to operate your Jinba?” dialog explains the scope; press “Allow” to let the AI list, create, and run Jinba flows on your behalf. This is a one-time, account-wide setting — you won’t be asked again in any chat, and the AI resumes the original request right away
• Choose “Not now” to skip. You can turn it on (or off) later from the “Let the AI create and edit workflows” band at the top of the Flow page’s list column
• If your organization’s settings disable AI code execution, the card shows “Unavailable — your organization’s settings disable AI code execution.” and cannot be enabled

​

Connect cards

When the conversation needs a service that is not connected yet — in a chat with or without an agent — the AI can offer to connect it right in the chat instead of sending you to the Connectors screen.

• Connector connect card — “Connect Gmail?” and the like for catalog connectors — Gmail, Google Calendar, Google Drive, Notion, Slack, Outlook, Microsoft Teams and, where your environment offers them, Google Docs, Google Sheets, OneDrive, SharePoint, and Excel. “Connect” opens the service’s sign-in flow in a popup, and the AI resumes your request once it completes (Personal connectors)
• Reconnect card — if a connected service stops working because its sign-in has expired or been revoked, the same card appears as “Reconnect …?” so you can restore the connection without leaving the chat
• Jinba Flow workspace connect card — “Connect Jinba Flow workspaces?” shows the organization-shared Jinba Flow workspaces you haven’t connected as a checklist; tick the ones you need and press “Connect.” In chats bound to an agent, you can also pick the Connection scope (“This agent only” / “Workspace-shared”) — that choice is offered only if you may edit the agent; otherwise the workspaces are connected workspace-shared (jinba-flow integration)

​

Agent hand-off card

Asking to talk to a specific agent — or describing work an existing agent is a better match for — can bring up, in a chat with or without an agent, an agent hand-off card (labeled “Call agent”) proposing that agent.

• The card shows the matched agent and an editable “First message to send” (up to 4,000 characters)
• “Start chat” opens a new chat bound to that agent; the first message is sent for you and the agent replies there. The original card keeps an “Open” link to the new chat
• “Dismiss” keeps you in the current chat

This is a hand-off, not an in-place switch — the chat you are in keeps its own agent setting (Selecting an agent).

​

Agent setting proposals

When you ask for it in chat — or when the AI spots something that would help — the AI can create a new agent or change the current agent’s setup. Each change is shown as a proposal card first, and nothing is applied until you approve it. If the current agent is shared with you as Use only — including an organization-shared agent opened outside its home workspace — its configuration is view-only. The AI explains that restriction instead of showing a settings, skill, connector, knowledge, or memory change proposal. The one exception is asking to register a file to Knowledge: on such an agent the AI offers the workspace save card instead, so the file goes into the workspace’s Knowledge rather than the agent’s. Ask the agent’s owner or a member with edit permission to make the change from the agent’s settings (Sharing agents).

• Agent creation proposed (in chats with no agent selected) — review the name, description, model, and system prompt, then press “Create.” An “Open” link takes you to the new agent
• Settings change proposed — shows what changes, including a system prompt diff. Press “Apply changes”; changes apply from the next chat
• Sharing change proposed — change who can use the agent (only you / everyone in this workspace / everyone in the organization) and whether they can also edit it. The card shows the change as before and after; unlike other settings, sharing takes effect immediately when you apply it
• Connector addition proposed — attach a connector already connected in the workspace to the agent. When several could fit, the card shows a picker grouped into “Jinba Flow workspaces” and “Custom MCP servers” — it opens even when nothing matches the name the AI mentioned, so just pick the right one. “Add” saves it right away — in most cases the card then reads “Ready to use in this chat”; otherwise it takes effect in new chats with this agent
• Skill attach proposed — attach a workspace-shared skill the workspace already has to the agent (Create Skills from Chat)
• Knowledge addition proposed — attach a knowledge source to the agent. Choose “Add” to attach it, or “Dismiss” to pass; additions take effect from your next chat
• Confirm memory action — turning the agent’s memory learning on or off; the switch flips only when you press “Run,” and the resolved card then states the outcome (“Turned on this agent’s memory” / “Turned off learning for this agent’s memory (existing entries remain and are still recalled)”). Erasing memory — single entries or all of it — is not possible from chat; the AI directs you to the agent’s Memory section instead (Agent memory)
• Confirm detach — removing an attached skill, connector, or knowledge item from the agent. Detaching saves right away; a shared item stays in the workspace catalog. Deleting the agent itself is not possible from chat (Managing agents)

For knowledge addition proposals, the caret on the “Add” button offers “Auto-approve the rest”: for the next 24 hours in that chat, further knowledge additions are applied without confirmation. While it is active, an “Auto-approving knowledge additions” banner above the composer reminds you (end it anytime with “Turn off”), and auto-approved cards carry an “auto” tag with an Undo link.

​

Task proposals

Ask for something to happen on a schedule — “run this every Monday morning,” “remind me next Friday” — and the AI answers with a “Task proposed” card (“Run this automatically on a schedule?” / “It will be saved as a Task”) instead of sending you to the Tasks page.

• The card shows the agent that will run, the title, the schedule in plain words with its timezone — “Daily 09:00 (Asia/Tokyo)” — or the date and time of a one-time run, the “Next run”, and the full “Instruction” the AI will follow on each run. When you named no timezone, a line says “Timezone taken from this device”
• “Create” saves it as a task that starts at its next run time; the card then reads “Created the task ”…"" with an “Open the task” link to the Tasks page. “Dismiss” passes. Nothing runs at the moment you approve
• Only creating works from chat. Editing, running now, pausing, and deleting an existing task are done on the Tasks page

​

Save-to-knowledge proposals

After producing a deliverable, the AI may offer to keep it for you with a “Save to knowledge proposed” card.

• The card spells out the name, the destination (a Knowledge folder, or the agent’s private knowledge), and who will be able to use the file. Press “Save” to register it, or “Dismiss” to pass
• “Change…” opens a picker where you can adjust the name and destination before saving
• Saving is create-only: if the name is already taken at the destination, the card warns you before you approve, and the file is saved under a new name like “name (2)” — nothing is ever overwritten
• These proposals always require an explicit click; the auto-approve mode above does not apply to them
• You can also save any deliverable yourself at any time — see Saving generated files

​

Email draft card

If Gmail is connected and you ask the AI to send an email, an email draft card appears in the chat. Review and edit the recipients, subject, and body before sending. You can also cancel with “Discard.” The AI never sends email without your confirmation. If your administrator has blocked Gmail for this workspace, pressing “Send via Gmail” shows “Your administrator has blocked Gmail sending in this workspace. You can still discard this draft.” on the card instead of sending.

​

Skill suggestions

When a reusable procedure takes shape during a chat, the AI may show a “Skill suggestion” card.

• Check what will be saved in the body preview
• Choose where to save it:
  • This agent only — usable only by the current agent (you can promote it to shared later)
  • Workspace shared — available to every agent
• Choose “Save” to register it, “Not now” to pass, or “Request changes” to have the AI revise the draft before saving (Create Skills from Chat)

When a request yields several skills at once, the AI shows one “Save as skills (N)” card instead: each draft is a row you can untick, the save location is chosen once for the whole card, and “Save n” creates the ticked rows together (Create Skills from Chat). If the AI detects that a conversation has drifted from an existing skill, it shows a “Skill update suggestion” — review the diff of the body and choose “Apply update.”