Chat Application
Chat is the interactive workspace for conversational agents, GeniAssistant, and supported built-in agents. It combines multimodal input, execution progress, tool-result rendering, user confirmation, session history, memory, and context management.
Open Chat
Launch Chat from the Platform Hub or open chat.genispace.ai. Choose an available agent from the sidebar. The list is filtered by the active space and your permissions.
Interface
| Area | Purpose |
|---|---|
| Agent sidebar | Select agents, switch sessions, and create a new conversation |
| Conversation | Read messages, progress events, confirmation cards, and plugin results |
| Input area | Enter text, upload supported content, choose mode, and view context usage |
| Activity Panel | Inspect tool arguments, structured results, errors, and plugin output |
The Activity Panel can be hidden, shown in a compact layout, or expanded depending on screen width.
Start and manage sessions
- Select New Chat to start with a clean conversation context.
- Open session history to continue an earlier conversation.
- Rename or delete a session from its menu when those actions are available.
- A resumed user-confirmation turn continues the same saved execution; it is not a new session.
Starting a new session does not automatically delete long-term memory.
Send a message
Text
Enter a request and press Enter or select Send. Use Shift+Enter for a new line.
Images
Drag supported images into the input area or use the image button. Image understanding depends on the selected model and agent configuration.
Documents
Attach supported documents through the file button. Upload status is shown before sending. Large or unsupported files may be rejected by the platform's file and model limits.
Voice
Use the microphone button when browser permission and speech input are available. Review the recognized text before sending.
Ask mode and Agent mode
Use the mode selector below the input:
- Ask — request a direct answer without delegated platform changes or multi-step tool execution.
- Agent — allow the agent to retrieve live information, call authorized tools, request confirmation, and complete multi-step work.
Agent mode does not bypass permissions. If you ask for a live search or resource change while using Ask mode, switch modes and send the request again.
Execution progress
Agent mode displays an execution timeline in the conversation. It may include retrieval, tool selection, tool execution, outcome checks, and response validation.
Select a tool event to open the detailed result in the Activity Panel. The timeline is an operational record and does not expose the model's private reasoning.
Tool results and plugins
Tool output may appear as:
- Structured JSON or a table
- A data-source or Dataset result
- A chart, preview, or generated resource summary
- A specialized Assistant plugin
- A confirmation form
The event summary and the detailed result serve different purposes. Open the event when you need to verify the returned records or error details.
Respond to a confirmation card
When the agent cannot safely continue without a subject, target, or choice, a confirmation card appears.
- Read the question and any selection limits.
- Select one or more options, or enter a custom answer in the provided input.
- Select Continue.
- The card closes and the saved turn resumes.
In single-choice mode, entering custom text is itself the selected answer; you do not also select an “Other” option.
Context usage and /compact
Long sessions can approach the model's context limit. When available, the input area shows the percentage used and remaining.
- GeniSpace can automatically summarize older context before the limit is reached.
- Enter
/compactto request compaction immediately. - The command does not send a normal message.
- Visible chat history remains available after compaction.
See Conversation Context Management.
Memory
If the selected agent has memory enabled, Chat can show memory controls for the current agent. Clearing memory requires confirmation. Conversation history, working context, and durable memory are separate; clearing one does not necessarily clear the others.
See Agent Memory.
Web search and retrieval
Agents may use web search, knowledge bases, datasets, data sources, and MCP tools when these capabilities are configured and authorized. Their availability depends on:
- Agent or built-in capability configuration
- Ask/Agent mode
- Current user and space permissions
- Data and tool access scopes
- Service availability
Interrupt, retry, and cancel
- Use Stop while a response is streaming to interrupt client output where supported.
- Retry a failed request after reviewing the error and tool details.
- A background job may continue after closing a dialog only when the interface explicitly says it continues in the background.
- A final chat response does not imply hidden tool work will continue.
GeniAssistant
Open the Assistant entry to ask platform questions or, in Agent mode, request supported platform operations. Describe the target resource, desired outcome, and constraints. GeniAssistant may request additional information before invoking a tool.
Accessibility and practical tips
- Use clear names or IDs when referring to an existing resource.
- State mandatory criteria separately from preferences.
- For Dataset retrieval, say whether you need an exact filter, literal keyword, or semantic match.
- Verify tool results before accepting recommendations that affect people, money, access, or production data.
- Start a new session when the topic changes completely.