v2.3

EduTelligence v2.3

Backfilled parity release for Artemis 9.3, whose release was published on 2026-05-25 21:31:14 UTC.

Compatibility

This release is compatible with Artemis 9.3.x.

Upgrade Notes

No EduTelligence breaking API or configuration change was identified in the EduTelligence PR descriptions for this release window.

Artemis 9.3 has an administrator-facing compatibility change: the Artemis athena Spring profile was replaced by the module feature flag pattern. Artemis deployments that enable Athena must set the corresponding artemis.athena.enabled: true flag when upgrading Artemis to 9.3.

Highlights

• Iris: model configuration can now opt into reasoning-effort support per model and explicitly declare temperature support. #549
• Athena: LLM modules better respect and log user AI preferences, and the LangChain/Python dependency stack was refreshed. #556, #559, #555, #564
• Logos: team management, application-key/team ownership changes, larger multimodal websocket frames, and capacity-planner stability fixes. #532, #571, #568, #569

Iris

Features

• Add per-model reasoning_effort support and explicit model capability flags by @Nayer-kotry in #549

Athena

Improvements

• Improve logging messages to include the user's AI preference by @maximiliansoelch in #556
• Respect LLM selection in module_text_llm by @maximiliansoelch in #559
• Migrate LangChain to v1.3 and update Python dependencies by @maximiliansoelch in #555
• Update Athena Poetry lockfiles by @maximiliansoelch in #564

Logos

Features

• Add team management, team ownership, application-key management, and user/team administration UI flows by @alex7sz in #532
• Improve the team-management UI and backing queries by @alex7sz in #571

Fixes and Operations

• Work around vLLM multimodal-cache sleep/wake desynchronization and broaden stuck-lane detection. direct commit c6be04f
• Exclude a wake target from its own eviction set to avoid stop-then-wake lane failures by @JakubJakubczyk in #563
• Raise websocket frame-size limits for large multimodal payloads by @wasnertobias in #568
• Fix a capacity-planner crash caused by missing e2e_latency_p50_seconds scheduler signals by @wasnertobias in #569
• Invalidate reflected metadata after applying Logos migrations. direct commit 95d14a5
• Add CUDA JIT header packages explicitly for the worker-node image. direct commit bf4dc85, direct commit d111b29

Full Changelog: v2.2...v2.3