Agent Desktop is the human and supervisory workspace inside Live Conversations. It brings active conversations, customer context, replies, assistance, actions, and operational controls into one place. Human agents can handle assigned work, while authorized supervisors can monitor ongoing AI-led and human conversations before deciding whether to intervene.
Agent Desktop is not only a transcript viewer or a separate AI copilot. It is the interface where people observe and operate customer conversations, collaborate with AI agents and external experts, and take over customer communication when human judgment or ownership is needed.
Find and organize active work
Depending on permissions and account configuration, the Live Conversations navigation can include:
- My Conversations for interactions the current human agent is handling.
- Help Needed for conversations where automation or another participant has requested human assistance.
- Recently Left for interactions the agent recently stopped handling.
- Reach Back for selected abandoned conversations that require follow-up.
- All In Progress and Overview for users who can supervise the wider operation.
Queues, filters, assignment, concurrency, and visibility depend on the account's routing and permission model.
Monitor, join, and take over AI-led conversations
Authorized supervisors can view conversations across All In Progress and the operational Overview, including conversations currently handled by automation without a human agent engaged. They can open an active conversation to inspect the transcript, participants, alerts, routing, and available context without first joining it.
When intervention is needed, a user with the required permissions can:
- join an ongoing conversation even when it has not entered Help Needed,
- take over customer communication from the AI-led workflow,
- assign the conversation to another human agent,
- contribute a bounded decision or a full customer-facing response,
- leave after the intervention so automation can continue when the configured workflow supports it.
Monitoring does not itself create a human-agent engagement. Joining does: the person becomes an active participant and can communicate with the customer. Whether automation pauses, remains silent, assists the person, or resumes afterward is determined by the configured workflow and operational controls.
Work with the customer conversation
The workspace can bring together:
- the live transcript and participant activity,
- customer identity and available customer-journey context,
- the reply composer and channel-specific fields,
- internal and external notes,
- files and attachments on supported channels,
- prepared replies and customer questions,
- information panels and links supplied for the current case,
- conversation attributes such as queue, priority, reason, and outcome.
The reply experience adapts to the channel. For example, email can include a subject, voice interactions can expose call controls, and supported messaging channels can offer their own delivery behavior. Confirm channel availability and limitations for the account before rollout.
Assistance while a human works
Depending on configuration, a human agent may receive:
- a suggested or prefilled reply that can be edited before sending,
- AI-assisted reply improvement,
- prepared replies and shortcuts,
- a quick message telling the customer that work is in progress,
- an agent-facing question that asks for a structured decision,
- information panels containing policy, case, customer, or system context,
- contextual links into an internal system or knowledge source.
These features reduce unnecessary navigation and help the human focus on the judgment or communication that requires a person.
Dynamic controls from automation logic
Configured automation logic can influence what the human sees and can do in Agent Desktop. This can include:
- proposing a default reply,
- showing, disabling, or hiding custom actions,
- showing or restricting conversation attributes,
- controlling standard actions such as leaving, resolving, waiting for the customer, asking for help, or inviting an external participant,
- displaying information panels,
- setting routing, priority, reason, outcome, customer classification, or an external case reference.
In Salted configuration and technical documentation, this automation logic is called Your Logic. The exact UI labels in configuration and reference documentation must remain unchanged. Concept pages explain the purpose; technical guides use the same names that users see in Salted.
Human-AI collaboration in Agent Desktop
A human does not always need to take permanent ownership of the interaction. Depending on the workflow, the human can:
- answer a bounded approval or guidance question,
- join and communicate with the customer,
- take over an unstructured or sensitive part of the case,
- ask another person for help,
- invite an external expert,
- leave after contributing the required judgment so AI can continue,
- proactively join and take over an AI-led conversation before automation explicitly requests help.
This preserves context and lets scarce human attention be applied where it changes the outcome.
What Agent Desktop does not decide by itself
Agent Desktop presents the conversation and controls. Business policy should still be enforced by the appropriate systems and automation logic. A consequential action should not depend only on a generated suggestion or a prompt when a deterministic rule, permission check, or human approval is required.
Technical documentation
This article explains the product concept. It does not replace configuration and reference guides.
- For the agent workflow, see Handling Live Conversation.
- For dynamic controls and suggested replies, see Dynamic Agent Desktop.
- For structured decisions shown only to agents, see Agent Facing Questions.
- For the broader operating model, see How AI and human agents work together in Salted.
Availability
The exact queues, channels, controls, information panels, actions, and AI assistance available in Agent Desktop depend on permissions, account configuration, integration, and current product maturity. Verify the relevant setup and reference documentation before making a production commitment.