v0.7.2 v0.8.0

Name: hindsight
Author: vectorize-io

high risk

vectorize-io/hindsightJun 2, 2026 → Jun 8, 20266 days between releases

What's New

Adds document transfer, mental model history, and prompt caching features.

New Features

Document transfer/export between banks
Mental model & observation history tables
LLM prompt-prefix caching
Claude Agent SDK integration
LangGraph support

Improvements

Per-bank LLM request tracing via OTel GenAI recorder
Pre-extraction freshness recheck and concurrent same-doc writer serialization
VectorChord BM25 gating and per-source candidate cap

Bug Fixes

Pre-extraction freshness recheck and concurrent same-doc writer serialization
VectorChord BM25 gating and per-source candidate cap
Persist bank config PATCH for never-retained banks
Localize operations and graph legends
Stop bank_id routing key polluting fact attribution

Upgrade Notes

Mental model history is now stored in dedicated tables, which may require schema updates. LLM prompt-prefix caching is enabled by default, altering prompt handling behavior.

The Two Releases

v0.7.2

Jun 2, 2026· 5 months ago

Focuses on new integrations (Vapi, Roo Code, Flowise, Google ADK) and performance optimizations, including Fireworks AI batch support.

Full release notes

v0.8.0

Jun 8, 2026· 5 months ago

A major release introducing document transfer/export, mental model history, LLM prompt caching, and new integrations like Claude Agent SDK and LangGraph.

Full release notes