v0.5.0

vectorize-io/hindsightv0.5.0Apr 8, 2026by github-actions[bot]

Upgrading?

Release notes summary

A major release focusing on performance improvements via a 3-phase retain pipeline, new integrations (OpenRouter, Google embeddings, llama.cpp), and security updates. It introduces new plugins and API features while removing legacy graph retrieval strategies.

Key Highlights

  • 3-phase retain pipeline to fix deadlocks and improve performance
  • New LLM providers: OpenRouter and built-in llama.cpp
  • New embeddings support: Google embeddings and reranker
  • Security updates including LiteLLM and Vite CVE fixes
  • Removal of BFS and MPFP graph retrieval strategies

Breaking Changes

  • remove BFS and MPFP graph retrieval strategies

New Features

  • Constellation view with Pretext canvas rendering
  • AutoGen integration for Hindsight
  • OpenCode persistent memory plugin
  • Bank template import/export
  • Retain update_mode='append' for document content concatenation
  • OpenRouter support for LLM, embeddings, and reranking
  • Built-in llama.cpp LLM provider for local inference
  • Hindsight-paperclip TypeScript integration

Full Release Notes

## What's Changed
* blog: What's New in Hindsight 0.4.22 by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/818
* Fix timeline group sort: localeCompare → numeric Date comparison by @ThePlenkov in https://github.com/vectorize-io/hindsight/pull/820
* fix(deps): address critical and high severity security vulnerabilities by @dcbouius in https://github.com/vectorize-io/hindsight/pull/827
* feat(control-plane): add Constellation view with Pretext canvas rendering by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/831
* fix(db): respect vector extension config in per-bank index migration by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/832
* breaking-change: remove BFS and MPFP graph retrieval strategies by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/767
* fix(consolidation): improve observation quality with structured processing rules by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/814
* perf: 3-phase retain pipeline — fix deadlocks, cap temporal links, query-time entity expansion by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/722
* Add OpenClaw shared memory team setup guide by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/788
* fix: resolve remaining Dependabot security alerts by @dcbouius in https://github.com/vectorize-io/hindsight/pull/833
* feat: add AutoGen integration for Hindsight by @DK09876 in https://github.com/vectorize-io/hindsight/pull/719
* fix: resolve 25 test regressions from streaming retain pipeline by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/836
* feat(openclaw): JSONL-backed retain queue for external API resilience by @akhater in https://github.com/vectorize-io/hindsight/pull/740
* fix(api): clear memories no longer deletes the bank by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/837
* fix(embed): clear stale daemon on port before starting (#843) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/845
* feat: add detail parameter to list/get mental models by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/846
* fix(retain): preserve normalized experience fact types by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/848
* feat(api): bank template import/export by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/819
* feat(paperclip): add hindsight-paperclip TypeScript integration by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/773
* Blog: Hindsight Is #1 on BEAM — the Benchmark That Tests Memory at 10M Tokens by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/851
* fix(ci): resolve all CI failures by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/847
* docs(hermes): update integration docs for plugin overhaul (hermes-agent#5094) by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/881
* blog: Hindsight is now a native memory provider in Hermes Agent by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/882
* blog: Persistent Memory for AutoGen Agents with Hindsight by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/883
* fix: use max_tokens for openai provider with custom base URL (Mistral compatibility) by @octo-patch in https://github.com/vectorize-io/hindsight/pull/858
* feat(helm): add persistent volume for local model cache by @isac322 in https://github.com/vectorize-io/hindsight/pull/861
* feat(google): add support for google embeddings and reranker by @ParadiseN1 in https://github.com/vectorize-io/hindsight/pull/863
* fix: Windows compatibility for hindsight-embed by @beordie in https://github.com/vectorize-io/hindsight/pull/867
* Fix AttributeError when event_date is None in fact_extraction by @Spectorisimo in https://github.com/vectorize-io/hindsight/pull/875
* chore(deps): bump actions/github-script from 7 to 8 by @dependabot[bot] in https://github.com/vectorize-io/hindsight/pull/879
* fix(query_analyzer): handle dateparser internal crashes gracefully by @cdbartholomew in https://github.com/vectorize-io/hindsight/pull/893
* fix(embed): skip profile .env overwrite when config has no HINDSIGHT_API_* keys by @octo-patch in https://github.com/vectorize-io/hindsight/pull/896
* fix(paperclip): address post-merge review fixes by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/900
* feat(recall): add proof_count boost to combined scoring by @Abdulkadirklc in https://github.com/vectorize-io/hindsight/pull/821
* fix(mcp): auto-coerce string-encoded JSON in tool arguments by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/899
* fix: make bank_id metric label opt-in to prevent OTel memory leak by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/898
* test: add regression tests for #874 and #894 by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/901
* feat: add OpenCode persistent memory plugin by @DK09876 in https://github.com/vectorize-io/hindsight/pull/853
* fix: add paperclip and opencode to changelog generator by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/903
* docs: add best practice for filtering recall by memory shape by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/905
* fix(config): validate entity_labels structure on PATCH by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/902
* fix(mcp): validate UUID inputs and add sync_retain tool by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/906
* security: bump litellm to >=1.83.0 (critical CVE fix) by @dcbouius in https://github.com/vectorize-io/hindsight/pull/912
* feat(openclaw): support bankId for static banks by @Aldoustheorchestrator in https://github.com/vectorize-io/hindsight/pull/910
* blog: One Memory for Every AI Tool I Use by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/914
* fix(recall): cap entity fanout in graph expansion by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/911
* fix: remove hindsight-cloud tag from guest post by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/916
* fix: update author name to Michael Bourkas by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/917
* fix: clamp out-of-range content_index in _map_results_to_contents by @mango766 in https://github.com/vectorize-io/hindsight/pull/908
* security: bump vite across integrations (high CVE fix) by @dcbouius in https://github.com/vectorize-io/hindsight/pull/913
* security: bump lodash, lodash-es, defu in root lockfile by @dcbouius in https://github.com/vectorize-io/hindsight/pull/915
* fix: make LiteLLM SDK embeddings encoding_format configurable by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/928
* feat: include occurred_end and mentioned_at in think-prompt fact serialization by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/929
* feat: add OpenRouter support for LLM, embeddings, and reranking by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/930
* chore: drop hindsight-hermes integration by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/931
* feat: add retain update_mode='append' for document content concatenation by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/932
* feat: add built-in llama.cpp LLM provider for local inference by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/933
* feat(openclaw): add configurable retain tags by @Aldoustheorchestrator in https://github.com/vectorize-io/hindsight/pull/937
* fix: exclude local-llm from [all] extra by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/936
* Add Codex persistent memory blog post by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/812

## New Contributors
* @isac322 made their first contribution in https://github.com/vectorize-io/hindsight/pull/861
* @Spectorisimo made their first contribution in https://github.com/vectorize-io/hindsight/pull/875
* @Abdulkadirklc made their first contribution in https://github.com/vectorize-io/hindsight/pull/821
* @Aldoustheorchestrator made their first contribution in https://github.com/vectorize-io/hindsight/pull/910
* @mango766 made their first contribution in https://github.com/vectorize-io/hindsight/pull/908

**Full Changelog**: https://github.com/vectorize-io/hindsight/compare/v0.4.22...v0.5.0

Related Repositories