Issues

• After a failed first build: consult uses the failed model, status says study never ran, next-step advice disagrees, consult cuts text silently

  bug
  blisspixel/deepr#223 · Filter by author blisspixel opened on Oct 14, 2026

• First-run leftovers: QUICK_START pip install -e, lenses vs chunks, init without TTY, relative data paths, stale review/currentness notes

  bug
  blisspixel/deepr#224 · Filter by author blisspixel opened on Oct 14, 2026

• Local consult synthesis overflows Ollama's default 4096 context on CPU/iGPU machines (study uses num_ctx, consult doesn't)

  bug
  blisspixel/deepr#222 · Filter by author blisspixel opened on Oct 14, 2026

• Knowledge view: show the excerpts each finding rests on and the stored support label for each position

  enhancement
  blisspixel/deepr#220 · Filter by author blisspixel opened on Oct 14, 2026

• Perspective: mark what rests on studied sources vs the expert's broader view, and stop rendering placeholder dissent as recorded

  enhancement
  blisspixel/deepr#221 · Filter by author blisspixel opened on Oct 14, 2026

• New Gemini users are steered to Gemini 2.5, which Google now limits to prior users

  bug
  blisspixel/deepr#217 · Filter by author blisspixel opened on Oct 13, 2026

• Local embedder: apply model-specific task prefixes, then evaluate EmbeddingGemma 2 through recall_quality

  blisspixel/deepr#216 · Filter by author blisspixel opened on Oct 13, 2026