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 #149: memory crash in Docker — missing vector index + pthread error

chopratejasaf3cd7f07fe4
diff --git a/headroom/memory/adapters/embedders.py b/headroom/memory/adapters/embedders.py
index 547b54da3..d04534e64 100644
--- a/headroom/memory/adapters/embedders.py
+++ b/headroom/memory/adapters/embedders.py
@@ -311,7 +311,13 @@ class OnnxLocalEmbedder:
model_path = hf_hub_download(self.ONNX_REPO, "model.onnx")
tok_path = hf_hub_download(self.ONNX_REPO, "tokenizer.json")
- self._session = ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])
+ # Set thread count to avoid pthread_setaffinity_np errors in Docker containers
+ sess_options = ort.SessionOptions()
+ sess_options.intra_op_num_threads = 1
+ sess_options.inter_op_num_threads = 1
+ self._session = ort.InferenceSession(
+ model_path, sess_options, providers=["CPUExecutionProvider"]
+ )
self._tokenizer = Tokenizer.from_file(tok_path)
self._tokenizer.enable_truncation(max_length=self._max_length)
self._tokenizer.enable_padding(length=self._max_length)
diff --git a/headroom/memory/factory.py b/headroom/memory/factory.py
index c90b27411..bbecd971a 100644
--- a/headroom/memory/factory.py
+++ b/headroom/memory/factory.py
@@ -148,14 +148,21 @@ def _create_vector_index(config: MemoryConfig) -> VectorIndex:
"""
backend = config.vector_backend
- # AUTO: prefer SQLITE_VEC if available, else HNSW
+ # AUTO: prefer SQLITE_VEC → HNSW → fail with helpful message
if backend == VectorBackend.AUTO:
- from headroom.memory.adapters import SQLITE_VEC_AVAILABLE
+ from headroom.memory.adapters import HNSW_AVAILABLE, SQLITE_VEC_AVAILABLE
if SQLITE_VEC_AVAILABLE:
backend = VectorBackend.SQLITE_VEC
- else:
+ elif HNSW_AVAILABLE:
backend = VectorBackend.HNSW
+ else:
+ raise ValueError(
+ "No vector index backend available for memory. Install one:\n"
+ " pip install sqlite-vec (recommended, lightweight)\n"
+ " pip install hnswlib (alternative)\n"
+ "Or install the full proxy bundle: pip install headroom-ai[proxy]"
+ )
if backend == VectorBackend.SQLITE_VEC:
from headroom.memory.adapters import SQLITE_VEC_AVAILABLE
diff --git a/pyproject.toml b/pyproject.toml
index 55b4c427f..e569d96fc 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -67,6 +67,7 @@ proxy = [
"onnxruntime>=1.16.0", # Kompress ONNX INT8 text compression (no torch needed)
"transformers>=4.30.0", # Tokenizer only (for Kompress)
"watchdog>=4.0.0", # File watcher for live code graph reindexing (--code-graph)
+ "sqlite-vec>=0.1.6", # Vector index for memory (--memory). Lightweight, no torch.
]
# AST-based code compression (tree-sitter)
code = [