v0.3.0

What's Changed

• Fix timing-unsafe bearer token comparison by @GSasu in #10
• Block SSRF in URL ingestion with DNS-level IP validation by @GSasu in #11
• Validate brainID to prevent path traversal in HTTP daemon by @GSasu in #12
• Fix pt store batch List to overlay journal state by @GSasu in #15
• Return error from http store New instead of panicking by @GSasu in #17
• Skip empty-id candidates in RRF fusion by @GSasu in #18
• Fix OpenAI embedder to include dimensions in API request by @GSasu in #13
• Handle tool_use content blocks in Anthropic streaming by @GSasu in #14
• Add HTTP client timeouts to all LLM providers by @GSasu in #16
• Fix TS dedup crash recovery — pipeline state tracking by @GSasu in #28
• Add prompt injection safety scanner with ML detection by @GSasu in #29
• Fix composite-concrete query detection to require both regexes by @GSasu in #19
• Strip markdown fences in TS LLM reranker JSON extraction by @GSasu in #21
• Port age-based heuristic confidence from TS to Go by @GSasu in #25
• Port parallel rerank batching from TS to Go by @GSasu in #23
• Port diversity-aware recall reranking from TS to Go by @GSasu in #22
• Port full episode management (CRUD/query) from TS to Go by @GSasu in #24
• Implement feedback classifier in TypeScript SDK by @GSasu in #26
• Add cost accounting to TypeScript eval framework by @GSasu in #27
• Align stale-superseded multiplier with TS implementation by @GSasu in #20

New Contributors

• @GSasu made their first contribution in #10

Full Changelog: v0.2.0...v0.3.0