Sibyl Tools

Custom tooling and patches for self-hosted Sibyl semantic memory deployments.

Contents

File	Lines	Purpose
lib/semantic_memory.py	358	Embedding layer — adds vector search via pplx-embed-v1-4b (OpenRouter) + sqlite-vec
bin/memory-add	35	CLI for adding memory entries
bin/memory-search-semantic	42	CLI for semantic search across memory
bin/migrate-holo-to-sibyl	137	One-way migration from holographic memory system to Sibyl
patch_sibyl.sh	789	Strips telemetry/auth/phoning-home from third-party Sibyl pip packages

Setup

# Install dependencies
pip install sibyl-memory-client sibyl-memory-mcp sibyl-memory-hermes sqlite-vec

# Apply patches (removes authorization/billing from the third-party packages)
bash patch_sibyl.sh

# Verify
python -c "from sibyl_memory_client._capcheck import TierCacheEntry; print('patched OK')"

Usage

# Add a memory
./bin/memory-add

# Semantic search
./bin/memory-search-semantic "what did we decide about the deployment?"

# Full migration
./bin/migrate-holo-to-sibyl

Design Decisions

• Direct SQLite access — the semantic_memory.py library reads/writes the Sibyl DB directly (state_documents and entities tables). No MCP dependency. ~2ms local queries.
• OpenRouter embeddings — uses pplx-embed-v1-4b via OpenRouter API. 2560-dimensional vectors stored in sqlite-vec tables.
• Patched client — patch_sibyl.sh replaces the upstream _capcheck.py (tier enforcement, network calls) with a no-op passthrough. All writes allowed, no phoning home, no billing checks.

License

MIT