AI Welcomes on Course Home
Course Home gives students a contextual next step, recent course updates, their questions, and their recent Iris conversations when those features are available. Its welcome works without AI: Artemis selects from 15 situations with three English and German variants each.
Enable a welcome provider
Home calls its own LLM preset, defined under artemis.llm like the presets of the other AI features (see LLM configuration). Enabling Iris, Atlas or Hyperion does not assign it. The feature is disabled by default. Assign a preset and its location to enable it:
artemis:
course-home:
ai:
enabled: true
location: LOCAL
llm-preset: welcome
Set location to LOCAL only when the preset's endpoint runs within the installation's approved local environment. Use CLOUD for an external provider. Students who selected local AI can receive welcomes only from a LOCAL provider. Students who selected cloud AI can use either location. Students who declined AI, have not selected a preference, or need to renew their consent receive deterministic welcomes.
When the Home provider is configured, students can open Account settings → AI experience, even if no other AI module is enabled. The local AI option in the selection dialog stays governed by info.localLLMDeploymentEnabled, which you set only when the connected services support local inference. A Home provider, even a local one, does not make that option available. Students can then receive Home welcomes through a local provider only after you enable that setting. Iris interaction counts and deletion controls appear only when Iris is enabled.
With enabled: true, the server does not start without llm-preset or with an unknown preset. A missing or unknown location leaves the provider unavailable, and deterministic welcomes continue to work. Earlier releases configured base-url, api-key and model here; move them to an endpoint and a preset. The endpoint's base URL must not contain credentials, query parameters or fragments; keep API keys in deployment secrets.
Request limits and fallback
The course overview loads independently of the model. Only an eligible welcome starts a separate request, with a three-second provider timeout, no retries, an output budget of 160 tokens (or the preset's lower limit), and at most four simultaneous requests per application instance. These limits apply whatever the preset's own timeout and retries are. Requests beyond that capacity fall back immediately. The client waits at most 3.5 seconds while continuing to show deterministic encouragement and usable course links.
Artemis rebuilds the authorized context and checks the student's current consent before dispatch and after the response. The bounded prompt contains published course description, instructor-authored learning guidance, the current situation, and a nonprivate next action. It excludes student identity, grades, question text, Iris conversation titles and transcripts, and source code. Private actions are described generically. Generated text is limited to encouragement; dates, counts, answer states and links come from course data. Invalid or unavailable model output keeps the deterministic welcome.
Course guidance and stored visits
Instructors can supply learning guidance in the course configuration. The student guide describes the resulting page.
Home records the first and most recent successfully displayed snapshot and a message variant cursor for each student and course. A read alone does not advance the visit. Course reset and permanent account deletion remove these records; personal data exports include them in course_home_visits.json. Home does not mark communication replies as read or create Iris conversations.