llmwiki-serve v0.2.10

llmwiki-serve 0.2.10

This release adds the opt-in SQLite GraphStore path for derived graph snapshots while keeping the default install and serving behavior backward-compatible.

Highlights:

• Optional SQLite GraphStore enabled at runtime with --graph-store sqlite --graph-store-path <outside-root.sqlite>.
• No new install extra for GraphStore; the base package uses Python's standard sqlite3 module.
• Existing [vector] and [redis] extras remain separate opt-ins for semantic retrieval and Redis/Valkey projection reuse.
• /graph, /graph/neighborhood, llmwiki_graph, and llmwiki_graph_neighbors preserve existing response contracts while using GraphStore when configured.
• Adds release gates and docs for visibility-scoped cache isolation, invalidation, fallback/fail-fast behavior, and future PostgreSQL SQL/PGQ re-evaluation.

Install:

uv tool install llmwiki-serve==0.2.10

SQLite GraphStore example:

llmwiki-serve serve ./wiki \
  --graph-store sqlite \
  --graph-store-path ../.llmwiki-cache/wiki-graph.sqlite

llmwiki-serve v0.2.9

llmwiki-serve 0.2.9

• Adds optional semantic retrieval preview through mode=vector and mode=hybrid across Python, HTTP, MCP, Streamable HTTP, and CLI query/search surfaces.
• Keeps lexical retrieval as the default path and documents agent-guided lexical retrieval as the recommended direct-agent workflow.
• Adds the optional llmwiki-serve[vector] extra with explicit local FastEmbed/NumPy bounds and operator-controlled model access.
• Adds deterministic chunking, cosine page ranking, fixed-k RRF hybrid ranking, bounded orientation hints, min-score rejection, capability metadata, and vector sidecar cache behavior without rewriting source orientation files.
• Preserves compatibility: no new endpoints, tools, commands, or result fields are required for existing lexical users.

llmwiki-serve 0.2.1

llmwiki-serve 0.2.1

• Adds optional Redis/Valkey projection-store support and release readiness documentation.
• Adds projection-store diagnostics fields for backend kind and sanitized Redis endpoint labels.
• Keeps in-memory projection as the default and treats Redis as opt-in derived cache storage.

Validation: CI passed, local Python gates passed, live Redis integration passed, release smoke and twine check passed.