AI Chat

AI Chat is the in-application assistant available from the tSM top bar on supported screens. It uses authenticated HTTP requests and Server-Sent Events (SSE) to stream progress and responses from the tsm-ai microservice. The deployment edge publishes /tsm-ai/api through tenant ingress or a configured tSM Gateway route and forwards it to the service facade.

Version note

tSM 2.4 introduced the top-bar chat and contextual SpEL integration. The unreleased 2.5 application adds specialized Form, TQL, and BPMN surfaces and prepares the common shell for contextual business scenarios. See the AI roadmap.

Access and Configuration

• The relevant AI capability must be enabled for the deployment.
• The user must have the privilege required by the active surface. SpEL AI uses Toolbar.AiChat; editor-specific surfaces may require the corresponding edit privilege.
• apiUrls.tsmAI configures the public tsm-ai base path. The default is /tsm-ai/api.
• The browser sends the current access token through the deployment edge. The authenticated REST/SSE facade validates the identity before accessing a conversation or assistant capability.

Supported Surfaces

Surface	Context and output
SpEL	Open script, script metadata, and selected microservice; returns explanations and a validated script proposal
Form	Open form/schema and form-designer context; returns explanations and a form proposal
TQL	Query, index/parameters, and selected microservice; returns an explained, compiled query proposal
BPMN	Open BPMN XML and process metadata; returns explanations and a validated BPMN proposal

The sparkle icon is shown only when the current screen has a supported surface and the user has the required access.

Using the Panel

Action	How
Open or close	Click the sparkle icon in the top bar
Send	Type a message and press Enter
New line	Press Shift+Enter
New conversation	Select New conversation in the panel
Resume a conversation	Open conversation history and select a thread
Stop a response	Use the stop action while a response is streaming

The panel can remain open while working with the associated editor. If the active artifact or microservice changes, check the visible context label before sending the next request.

Attachments

AI Chat accepts up to five attachments, with a maximum of 5 MiB per file and 10 MiB in total. Supported categories include images, PDF, text, and selected source files. Images can also be pasted from the clipboard.

Attachment content is sent with the current request. Conversation history keeps attachment metadata, not a reusable copy of the original binary. Reattach a file when a later request must analyze its contents again.

Streaming Responses

Responses arrive as structured SSE events. The panel can show:

• progress and status steps;
• streamed Markdown text and syntax-highlighted code;
• validation or execution result metadata;
• one or more typed artifact suggestions;
• a terminal success or sanitized error state.

Progress is operational status, not a stored private chain-of-thought trace.

Reviewing Editor Changes

For an editor surface, AI Chat returns a complete proposed artifact plus an explanation. The normal flow is:

1. The assistant validates the proposal using the surface-specific deterministic gate.
2. tSM displays a diff or review action against the current editor content.
3. Select Accept to apply the proposal to the editor or Reject to keep the current artifact.
4. Save, test, publish, or deploy through the normal configuration lifecycle.

Accepting a suggestion changes the editor draft; it does not bypass validation, Change Sets, packages, privileges, or deployment controls.

Conversation History

When the tsm-ai conversation store is configured, threads are stored server-side and authorized by user, tenant, and assistant surface. Deployments using local browser history treat it as personal ephemeral state rather than a shared business record.

Threads are surface-specific. A SpEL thread is not automatically reused as a BPMN, Form, or TQL thread because the context and artifact contracts differ.

Feedback

Use thumbs up or down on an assistant response. Negative feedback can include a score or comment. Feedback is linked to the request/response and assistant surface so it can be used for evaluation and quality improvement.

Security Notes

• AI Chat acts within the authenticated tSM session and the configured assistant capabilities.
• The model does not choose the tenant host or receive arbitrary backend credentials.
• Editor validation and optional read-only execution use server-controlled routes.
• Do not paste secrets or data that is not approved for the deployment's model provider.
• A generated artifact or future business action remains a proposal until the normal tSM control explicitly applies or executes it.

Planned Business Context

The 2.5 roadmap extends the same shell to customer, ticket, order, listing, and other business screens through versioned Assistant Profiles and server-side context providers. Planned write actions use typed proposals, preview/confirmation, reauthorization, and — for long-running or high-impact work — a durable process. Process-initiated AI runs execute in the Agent Runtime module of the same tsm-ai microservice and are correlated with durable Process Engine wait states.

See:

• Contextual AI Assistant Architecture
• tSM AI and Agent Runtime
• AI Smart Operations
• AI Assisted Development