v1.5.6 — GPT4All, vLLM, SQLiteVecStore, 4 new loaders

What's new

Added

• GPT4AllLLM — local model provider via GPT4All Python bindings; no API key required; pip install synapsekit[gpt4all]
• VLLMLlm — high-throughput local/self-hosted inference via vLLM's OpenAI-compatible API; pip install synapsekit[vllm]
• SQLiteVecStore — zero-infra vector store backed by sqlite-vec; drop-in for InMemoryVectorStore; pip install synapsekit[sqlite-vec]
• ParquetLoader — load Parquet files as Documents; pip install synapsekit[parquet]
• RedisLoader — load key/value pairs from Redis (string, hash, JSON); pip install synapsekit[redis]
• ElasticsearchLoader — load from an Elasticsearch index; search + scan modes; pip install synapsekit[elasticsearch]
• DynamoDBLoader — load from AWS DynamoDB; scan/query with auto-pagination; pip install synapsekit[dynamodb]
• Production-grade test suite — 120 new tests: preflight, E2E, behavioral, API endpoint

Fixed

• Stream disconnect race condition — client disconnects now terminate cleanly
• Summary buffer memory corruption — buffer no longer mutated before LLM summarisation completes

Counts

• 33 LLM providers (was 31)
• 46 loaders (was 37 at v1.5.5 start, 41 after batch merge)
• 10 vector stores (was 9, +SQLiteVecStore)

Install

pip install synapsekit==1.5.6

Full changelog: https://github.com/SynapseKit/SynapseKit/blob/main/CHANGELOG.md