feat(sbx,ai): snapshot all backends w/ local export; robust AI + rejoin

- /sbx save [--local] and new /sbx vmsave/vmsnaps: snapshot Docker,
  Multipass, and VirtualBox; --local also writes a portable artifact
  (docker .tar / VBox .ova) under hh-snapshots/ that survives pruning.
- sandbox reappears for anyone who leaves and rejoins (host replays
  status + screen snapshot + ACL on Joined); SbxStatus ready handler is
  now idempotent so it never wipes scrollback.
- received files auto-bridge into the hosted sandbox (ft::tar_path).
- AI agent: translate Ollama's cryptic 404 into model/host/fix guidance.
- bootstrap installs the AI layer (Ollama + default model) by default,
  with consent gates; --no-ai opts out, --yes skips prompts.
- help menu lists the new save/vm flags.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

bootstrap.sh

@@ -5,11 +5,18 @@
 # for the server, installs its dependencies, and builds the Rust client.
 #
 # usage:
-#   ./bootstrap.sh              # full setup (venv + deps + cargo build)
+#   ./bootstrap.sh              # full setup (venv + deps + client + AI layer)
 #   ./bootstrap.sh --release    # build the client in release mode
+#   ./bootstrap.sh --no-ai      # skip the AI layer (no Ollama install / model pull)
+#   ./bootstrap.sh --yes        # don't prompt during the AI layer (install / pull)
 #   ./bootstrap.sh --check      # only report what's installed; change nothing
 #   ./bootstrap.sh -h | --help  # this help
 #
+# The AI layer (local Ollama + a default model for the /ai agent) is set up by
+# default so the agent works out of the box and nobody hits "model not found".
+# It still prompts before installing/pulling on an interactive terminal — skip
+# the whole thing with --no-ai, or skip the prompts with --yes.
+#
 # After it finishes, spin up a local test session with:  cd hh && ./lets-hack.sh
 set -uo pipefail
 
@@ -19,12 +26,16 @@ HH_DIR="$ROOT/hh"
 
 RELEASE=0
 CHECK_ONLY=0
+DO_AI=1
+ASSUME_YES=0
 for arg in "$@"; do
     case "$arg" in
         --release) RELEASE=1 ;;
         --check)   CHECK_ONLY=1 ;;
+        --no-ai)   DO_AI=0 ;;
+        --yes|-y)  ASSUME_YES=1 ;;
         -h|--help|-help) grep '^#' "$0" | sed 's/^# \{0,1\}//'; exit 0 ;;
-        *) echo "✖ unknown arg: $arg (try --release / --check / --help)" >&2; exit 2 ;;
+        *) echo "✖ unknown arg: $arg (try --release / --no-ai / --yes / --check / --help)" >&2; exit 2 ;;
     esac
 done
 
@@ -73,8 +84,22 @@ else
     echo "  ✓ client built: hh/target/debug/hack-house"
 fi
 
+# 4. AI layer (on by default): install Ollama + pull the default model so the
+#    /ai agent works out of the box and nobody hits "model not found" later. The
+#    logic lives in bootstrap-ai.sh; --ai-only skips its baseline re-run (we just
+#    did it). Declining/failing here is non-fatal — the baseline setup still
+#    stands and the AI layer can be added later with ./bootstrap-ai.sh.
+if [[ $DO_AI -eq 1 ]]; then
+    ai_args=(--ai-only)
+    [[ $ASSUME_YES -eq 1 ]] && ai_args+=(--yes)
+    "$ROOT/bootstrap-ai.sh" "${ai_args[@]}" \
+        || echo "⚠ AI layer not completed — re-run ./bootstrap-ai.sh when ready" >&2
+fi
+
 echo
 echo "ready. next steps:"
 echo "    cd hh && ./lets-hack.sh          # local test session (server + clients in tmux)"
 echo "    # or run the server + client by hand — see README.MD"
-echo "    # want the local AI agent? ./bootstrap-ai.sh  (installs Ollama + a model)"
+if [[ $DO_AI -eq 0 ]]; then
+    echo "    # AI layer skipped (--no-ai); add it later with ./bootstrap-ai.sh"
+fi