Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

fix(memory): use ONNX embedder for `wrap --memory` sync (#1092) (#1262)

Parideboy4f9fedaa7a02
diff --git a/headroom/cli/wrap.py b/headroom/cli/wrap.py
index 4335940de..4b5ef2f1b 100644
--- a/headroom/cli/wrap.py
+++ b/headroom/cli/wrap.py
@@ -3719,8 +3719,7 @@ def codex(
try:
import asyncio
- from headroom.memory.backends.local import LocalBackend, LocalBackendConfig
- from headroom.memory.sync import sync_import
+ from headroom.memory.sync import _build_sync_backend, sync_import
from headroom.memory.sync_adapters.claude_code import (
ClaudeCodeAdapter,
get_claude_memory_dir,
@@ -3729,8 +3728,7 @@ def codex(
claude_memory_dir = get_claude_memory_dir()
async def _import_claude_memories() -> int:
- config = LocalBackendConfig(db_path=db_path)
- backend = LocalBackend(config)
+ backend = _build_sync_backend(db_path)
await backend._ensure_initialized()
adapter = ClaudeCodeAdapter(claude_memory_dir)
count = await sync_import(backend, adapter, mem_user)
diff --git a/headroom/memory/sync.py b/headroom/memory/sync.py
index 8f78e12c5..ce4f0c596 100644
--- a/headroom/memory/sync.py
+++ b/headroom/memory/sync.py
@@ -343,6 +343,21 @@ async def sync_export(
# ---------------------------------------------------------------------------
+def _build_sync_backend(db_path: str) -> Any:
+ """Build the memory backend used by the sync subprocess.
+
+ Match the proxy MCP server (see ``headroom/memory/mcp_server.py``): use the
+ torch-free ONNX embedder so ``wrap --memory`` sync works on the proxy extras
+ without sentence-transformers/PyTorch (#1092). It loads the same
+ ``all-MiniLM-L6-v2`` 384-dim model as the local embedder, so vectors stay
+ compatible with what the proxy writes — no DB migration.
+ """
+ from headroom.memory.backends.local import LocalBackend, LocalBackendConfig
+
+ config = LocalBackendConfig(db_path=db_path, embedder_backend="onnx")
+ return LocalBackend(config)
+
+
def main() -> None:
"""CLI entry point for running sync from a subprocess."""
import argparse
@@ -358,10 +373,7 @@ def main() -> None:
import json as _json
async def _run() -> None:
- from headroom.memory.backends.local import LocalBackend, LocalBackendConfig
-
- config = LocalBackendConfig(db_path=args.db)
- backend = LocalBackend(config)
+ backend = _build_sync_backend(args.db)
await backend._ensure_initialized()
if args.agent == "claude":
diff --git a/tests/test_memory_sync.py b/tests/test_memory_sync.py
index 903c63915..3ff0f0346 100644
--- a/tests/test_memory_sync.py
+++ b/tests/test_memory_sync.py
@@ -23,6 +23,7 @@ from typing import Any
import pytest
from headroom.memory.sync import (
+ _build_sync_backend,
sync,
sync_export,
sync_import,
@@ -675,3 +676,14 @@ class TestCrossAgentInterop:
# AGENTS.md has both (from DB)
agents_content = agents_md.read_text()
assert "FastAPI" in agents_content or "8787" in agents_content
+
+
+def test_sync_backend_uses_onnx_embedder(tmp_path):
+ """#1092: the sync subprocess must pick the torch-free ONNX embedder.
+
+ Defaulting to the LOCAL (sentence-transformers) embedder makes
+ `wrap --memory` crash with an ImportError on the proxy extras. The backend
+ must match the proxy MCP server, which uses ONNX.
+ """
+ backend = _build_sync_backend(str(tmp_path / "memory.db"))
+ assert backend._config.embedder_backend == "onnx"