Release v0.8.0 — local LLMs, image input, the ingestlib CLI, slim installs

Highlights since 0.7.0:

- BEHAVIOR CHANGE: the default vector_store is now sqlite (was pinecone).
  Existing configs that relied on the old default must set
  `vector_store: pinecone` explicitly.
- Ollama provider: run every LLM and embedding call on your own machine
  (llm_provider/embedding_provider: ollama; reference stack qwen3.5:9b +
  qwen3-embedding:0.6b, GGUF builds).
- Image input: PNG/JPEG/WebP parse as one-page documents; classify reads
  them via vision; ingest accepts them end to end.
- The ingestlib CLI: `init` scaffolds config.yaml/.env (--local writes the
  zero-cloud preset), `doctor` verifies every configured choice with real
  calls and one-sentence fixes, `--version`.
- Slim installs: connector SDKs are pip extras (ingestlib[pinecone|qdrant|
  pgvector|mongodb|milvus|opensearch|weaviate|all]); ~1.3 GB core covers
  the full default stack; storage imports connectors lazily.
- The aws config section is now required only when a choice uses AWS.
- Error hints at every backend boundary: wrong AWS profile lists your
  available profiles, dead Ollama says `ollama serve`, exhausted Jina
  quota names the dashboard, missing SDK prints its pip extra.
- Docs rebuilt: langmodule.github.io/ingestlib — journey-structured
  quickstart, concepts, task guides, troubleshooting catalog, full
  configuration reference.
- ingestlib.__version__, python -m ingestlib, notebook-friendly sync
  wrappers, quieter third-party logging.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: ramacharanreddy-k

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ingestlib"
-version = "0.7.0"
+version = "0.8.0"
 description = "Self-hosted document intelligence for RAG — parse, classify, and split PDF/DOCX/PPTX and images into searchable, cited, retrieval-ready chunks"
 readme = "README.md"
 license = "MIT"