User companion

DelphAI xLM

Standalone user-facing runtime for chat, saved prompts, schedule-driven briefings, and league-grounded sports assistance.

History surface

Recover prior work without losing context

History is not a transcript dump. It is the place where users revisit saved briefings, understand what the companion knew at the time, and restart from high-value conversations.

Storage split

Operational turns are short-lived and query-heavy. Curated prompt assets, references, and generated reports belong in PostgreSQL context storage.

Recall mode

Persisted turns

Recent conversation list loaded from companion.inference_run.

Retention split

Operational + curated

Operational turns and curated prompt assets should not collapse into one store.

Blocked state

Explicit

History preserves the actual model and response text returned during grounded runs.

Open route

Open live session

Jump into a mock conversation with game and schedule context.

Open route

Inspect history

See the report and recall surfaces this runtime has to support.

Open route

Review prompt catalog

Ground reusable prompts before orchestration and persistence wire up.

Saved sessions

Users need recall, restart, and promotion paths

These entries are the first concrete contract for list views, search, pinning, export state, and capability disclosures.