diff --git a/src/gen_worker/cozy_cas.py b/src/gen_worker/cozy_cas.py
index 6866c89..45836a0 100644
--- a/src/gen_worker/cozy_cas.py
+++ b/src/gen_worker/cozy_cas.py
@@ -324,7 +324,7 @@ async def _download_one_file(url: str, dst: Path, expected_size: int, expected_b
     # are not killed.  total=None lets multi-GB downloads run as long as data
     # keeps flowing; sock_read=120 catches genuine stalls.
     timeout = aiohttp.ClientTimeout(total=None, sock_connect=float(os.getenv("WORKER_MODEL_DOWNLOAD_SOCK_CONNECT_TIMEOUT_S", "60")),
-                                                                    sock_read=float(os.getenv("WORKER_MODEL_DOWNLOAD_SOCK_READ_TIMEOUT_S", "120")))
+                                                                    sock_read=float(os.getenv("WORKER_MODEL_DOWNLOAD_SOCK_READ_TIMEOUT_S", "180")))
     tmp = dst.with_suffix(dst.suffix + ".part")
     # If we have a partial file, try to resume via HTTP Range.
     offset = 0
diff --git a/src/gen_worker/cozy_snapshot_v2_downloader.py b/src/gen_worker/cozy_snapshot_v2_downloader.py
index cd74459..9399988 100644
--- a/src/gen_worker/cozy_snapshot_v2_downloader.py
+++ b/src/gen_worker/cozy_snapshot_v2_downloader.py
@@ -1,7 +1,9 @@
 from __future__ import annotations
 
 import asyncio
+import json
 import os
+import re
 import shutil
 import threading
 from pathlib import Path
@@ -21,6 +23,39 @@ def _blob_path(blobs_root: Path, digest: str) -> Path:
     return blobs_root / "blake3" / digest[:2] / digest[2:4] / digest
 
 
+_PART_FILE_RE = re.compile(r"\.part\d{4}$")
+
+
+def _strip_blake3_prefix(digest: str) -> str:
+    """Strip the 'blake3:' scheme prefix if present, returning the bare hex."""
+    d = (digest or "").strip().lower()
+    if d.startswith("blake3:"):
+        d = d[len("blake3:"):]
+    return d
+
+
+def _is_part_file(path: str) -> bool:
+    """Return True if the path is a chunked part file (e.g. foo.part0001)."""
+    return bool(_PART_FILE_RE.search(path))
+
+
+def _is_parts_manifest(path: str) -> bool:
+    """Return True if the path is a chunked-blob manifes (e.g. foo.parts.json)."""
+    return path.endswith(".parts.json")
+
+
+def _resolve_field(obj: Any, *keys: str) -> Any:
+    """Get a field from either a dict or an attribute-bearing object, trying keys in order."""
+    for k in keys:
+        if isinstance(obj, dict):
+            v = obj.get(k)
+        else:
+            v = getattr(obj, k, None)
+        if v is not None:
+            return v
+    return None
+
+
 def _try_hardlink_or_copy(src: Path, dst: Path) -> None:
     if dst.exists():
         dst.unlink()
@@ -38,24 +73,34 @@ def _try_hardlink_or_copy(src: Path, dst: Path) -> None:
 
 
 def _coerce_resolved_model(ref: CozyRef, resolved: Any) -> CozyHubResolveArtifactResult:
-    """Coerce an orchestrator-provided resolved model object into CozyHubResolveArtifactResult."""
-    snapshot_digest = str(getattr(resolved, "snapshot_digest", "") or "").strip()
-    if not snapshot_digest:
-        snapshot_digest = str(getattr(resolved, "snapshotDigest", "") or "").strip()
+    """Coerce an orchestrator-provided resolved model object into CozyHubResolveArtifactResult.
+
+    Handles two wire shapes:
+      - Legacy:  .snapshot_digest (bare hex) + .files[]
+      - v2:      .snapshot_digest ('blake3:...' prefixed) + .entries[]  (new chunked-blob format)
+
+    Both protobuf attribute access and plain-dict access are supported.
+    """
+    snapshot_digest = str(_resolve_field(resolved, "snapshot_digest", "snapshotDigest") or "").strip()
     if not snapshot_digest:
         raise ValueError("resolved model missing snapshot_digest")
+    # Strip scheme prefix so the digest is a bare hex string suitable for path use.
+    snapshot_digest = _strip_blake3_prefix(snapshot_digest) or snapshot_digest
 
-    files_raw = list(getattr(resolved, "files", []) or [])
+    # New format uses "entries"; legacy format uses "files".
+    files_raw = list(_resolve_field(resolved, "entries", "files") or [])
     files: List[CozyHubSnapshotFile] = []
     for ent in files_raw:
-        path = str(getattr(ent, "path", "") or "").strip()
+        path = str(_resolve_field(ent, "path") or "").strip()
         if not path:
             continue
-        blake3_hex = str(getattr(ent, "blake3", "") or "").strip().lower()
+        # Prefer bare "blake3" field; fall back to "digest" which may carry the prefix.
+        blake3_hex = str(_resolve_field(ent, "blake3", "BLAKE3") or "").strip().lower()
         if not blake3_hex:
-            blake3_hex = str(getattr(ent, "BLAKE3", "") or "").strip().lower()
-        url = str(getattr(ent, "url", "") or "").strip() or None
-        size_bytes = int(getattr(ent, "size_bytes", 0) or 0)
+            digest_raw = str(_resolve_field(ent, "digest") or "").strip().lower()
+            blake3_hex = _strip_blake3_prefix(digest_raw)
+        url = str(_resolve_field(ent, "url") or "").strip() or None
+        size_bytes = int(_resolve_field(ent, "size_bytes") or 0)
         if not blake3_hex or not url:
             raise ValueError(f"resolved model file missing blake3/url: {path}")
         files.append(CozyHubSnapshotFile(path=path, size_bytes=size_bytes, blake3=blake3_hex, url=url))
@@ -123,7 +168,50 @@ async def ensure_snapshot(
 
             tmp = snaps_root / f"{res.snapshot_digest}.building"
             tmp.mkdir(parents=True, exist_ok=True)
+
+            # Reassemble any chunked files (produced by chunkedblob on ingest).
+            # A ".parts.json" entry describes how to reassemble N part blobs into
+            # the original file.  Part blobs and the manifest itself are all already
+            # in the blob store at this point.
+            parts_manifest_entries = [f for f in res.files if _is_parts_manifest(f.path)]
+            part_file_paths = {f.path for f in res.files if _is_part_file(f.path)}
+
+            for pm_entry in parts_manifest_entries:
+                parts_json_blob = _blob_path(blobs_root, pm_entry.blake3)
+                try:
+                    parts_manifest = json.loads(parts_json_blob.read_bytes())
+                except Exception as exc:
+                    raise ValueError(f"failed to parse parts manifest {pm_entry.path}: {exc}") from exc
+
+                original_path = str(parts_manifest.get("original_path") or "").strip()
+                if not original_path:
+                    raise ValueError(f"parts manifest {pm_entry.path} missing original_path")
+                parts = parts_manifest.get("parts") or []
+                if not parts:
+                    raise ValueError(f"parts manifest {pm_entry.path} has no parts")
+
+                rel = _norm_rel_path(original_path)
+                dst = tmp / rel
+                dst.parent.mkdir(parents=True, exist_ok=True)
+                if dst.exists():
+                    dst.unlink()
+
+                with open(dst, "wb") as out_f:
+                    for part in parts:
+                        part_digest = _strip_blake3_prefix(
+                            str(part.get("digest") or "").strip().lower()
+                        )
+                        if not part_digest:
+                            raise ValueError(f"part entry in {pm_entry.path} missing digest")
+                        part_blob = _blob_path(blobs_root, part_digest)
+                        with open(part_blob, "rb") as in_f:
+                            shutil.copyfileobj(in_f, out_f)
+
+            # Materialize regular files; skip part files and parts manifests since they
+            # have already been consumed above during reassembly.
             for f in res.files:
+                if _is_parts_manifest(f.path) or f.path in part_file_paths:
+                    continue
                 rel = _norm_rel_path(f.path)
                 dst = tmp / rel
                 dst.parent.mkdir(parents=True, exist_ok=True)
diff --git a/uv.lock b/uv.lock
index 7db9fb9..056c247 100644
--- a/uv.lock
+++ b/uv.lock
@@ -671,7 +671,7 @@ wheels = [
 
 [[package]]
 name = "gen-worker"
-version = "0.3.8"
+version = "0.3.9"
 source = { editable = "." }
 dependencies = [
     { name = "aiohttp" },