Hindsight Files

What's Changed

• docs: 0.9.0 changelog + blog posts by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3189
• fix(coding-agents): seed the actual harness, not the "opencode" default (#3247, [#3248]) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3266
• fix(docker): repair PGroonga Compose image by @JiehoonKwak in https://github.com/vectorize-io/hindsight/pull/3316
• fix(engine): stop concurrent bank deletes from deadlocking on vector-index DDL by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3245
• fix(coding-agents): fall back from filesystem-root worktrees by @koriyoshi2041 in https://github.com/vectorize-io/hindsight/pull/3286
• fix(worker): serialise graph_maintenance per bank at claim time by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3235
• fix(llm): forward extra body to Codex requests by @koriyoshi2041 in https://github.com/vectorize-io/hindsight/pull/3305
• fix(api): knowledge-base search 500s on non-native text-search backends (#3268) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3318
• test(ci): gate every build on hermes-agent@main co-installability by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3265
• fix(config): validate bank config value types and stop them wedging tasks (#3218) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3319
• test(bank-template): round-trip every exportable field through export+import by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3324
• perf(reflect): drop retrieval plumbing from reflect tool results by @JoshFunnell in https://github.com/vectorize-io/hindsight/pull/3310
• feat(bank-template): make every bank config field export+importable by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3332
• fix(structured-doc): add TableBlock to fix markdown table rendering by @dragonkid in https://github.com/vectorize-io/hindsight/pull/3289
• feat(reflect): resolve entity names on reflect sub-recalls by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3334
• fix(api): gate knowledge-base routes through the operation validator (#3312) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3331
• fix(coding-agents): write back only the new turns (append + idempotent retain) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3336
• Pin embed to the metapackage release by @koriyoshi2041 in https://github.com/vectorize-io/hindsight/pull/3261
• feat(transfer): carry Knowledge Pages tree + regenerate mental-model search state on import (#3308, [#3323]) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3330
• fix(api): return 400 for malformed UUID on GET /entities/{id} and /memories/{id}/history by @Alan5168 in https://github.com/vectorize-io/hindsight/pull/3260
• fix(control-plane): surface direct request failures by @Sanderhoff-alt in https://github.com/vectorize-io/hindsight/pull/3333
• fix(api): add a DB-free liveness probe so a slow database stops restarting pods by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3337
• fix(coding-agents,ci): drop the periodic re-sync; stop running doc examples for integration changes by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3341
• fix(coding-agents): bound transcript reads so an oversized session still retains (#3292) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3345
• fix(ollama): make native think configurable via extra_body by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3344
• feat(llm): xai-oauth provider — run the engine's LLM lanes on a flat-rate SuperGrok subscription (device-code OAuth, no API key), production-hardened by @JoshFunnell in https://github.com/vectorize-io/hindsight/pull/3272
• fix(reflect): provide the current date for temporal reasoning by @koriyoshi2041 in https://github.com/vectorize-io/hindsight/pull/3287
• fix(transfer): reject a wrong import zip with a 400 that names the fix by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3339
• blog: guest post — writenode, continuity over retrieval (Josh Groves) by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3243
• docs(config): name LM Studio/Ollama/Volcano for LLM_STRICT_SCHEMA by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3348
• fix(search): use PGroonga for Knowledge Pages by @JiehoonKwak in https://github.com/vectorize-io/hindsight/pull/3335
• fix(engine): normalize whitespace in candidate entity names at resolution intake by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3338
• fix(gemini): fail fast on deterministic 400 INVALID_ARGUMENT (#3256) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3347
• fix(ci): compare OpenAPI against the merge-base; build benchmark role configs whole by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3349
• feat(coding-agents): retainTags / retainMetadata, with HINDSIGHT_RETAIN_TAGS (#3269, [#2896]) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3346
• fix(coding-agents): seed bank missions once, then leave them to the user (#2492) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3352
• fix(coding-agents): report configured MCP harness by @TheAngryPit in https://github.com/vectorize-io/hindsight/pull/3342
• feat(llm): restore server-side prompt caching on load-balanced OpenAI-compatible backends — measured 0.085% aggregate hit rate without affinity vs 99.4% with, same backend by @JoshFunnell in https://github.com/vectorize-io/hindsight/pull/3271
• fix(retain): cut oversized sub-batches on native chunk boundaries (#3282) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3351
• perf(docs): switch the Docusaurus build to Rspack + SWC by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3357
• feat(transfer): async document export (#3321) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3340
• fix(obsidian): scope sync index to bank and API target (#3257) by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3354
• fix(coding-agents): drop harness transport wrappers from retained turns (#3023) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3378
• docs(coding-agents): say which harnesses retainEveryTurns applies to by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3380
• fix(transfer): mint a fresh internal_id on whole-bank import (#3270) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3353
• fix(coding-agents): drop Claude Code's compaction summary from retained turns (#3379) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3383
• fix deletion of failed document uploads by @koriyoshi2041 in https://github.com/vectorize-io/hindsight/pull/3366
• fix(retain): sweep observations when delta retain deletes chunks by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3384
• docs: correct the Azure OpenAI base URL by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3385
• feat(llm): opt-in forced-tool structured output for LiteLLM providers (#3300) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3382
• feat(memories): per-bank store capabilities on main (writes_memory_rows_in_sql_for / owns_document_store_for) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3388
• fix(retain): fold shared document_id items on the sync path (#3363) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3386
• fix(reflect): decouple page max_tokens from the provider output cap (#3365) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3389
• fix(retain): match chunk-delete link endpoints through indexable joins (#3387) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3393
• feat(agent-plugin): portable Hindsight plugin for the Agent Plugins standard by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3394
• fix(retain): stop concurrent appends to one document from losing turns by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3395
• fix(stats): drop permanently failed memories from pending_consolidation by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3397
• blog: Best Open-Source Agent Memory Systems (Self-Hosted, 2026) by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3192
• blog: update Best Open-Source Agent Memory Systems date to 2026-08-11 by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3399
• fix(reflect): split synthesis — parallel claim extraction + reduce instead of dropping over-budget evidence by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3392
• fix(coding-agents): bound automatic reflect budget by @TheAngryPit in https://github.com/vectorize-io/hindsight/pull/3364
• feat(coding-agents): add Prime Agent as a supported harness by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3240
• fix(compose): use HINDSIGHT_API_LLM_API_KEY env var by @ishfuseini in https://github.com/vectorize-io/hindsight/pull/3398
• fix(coding-agents): resolve the project when the working directory is gone (supersedes [#3110]) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3410
• fix(retain): globalise memory_links lock order on the insert path (#3396) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3406
• perf(graph): decouple stale-cooccurrence prune from hub-entity degree (#3367) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3408
• docs(reflect): clarify tag-scoped directive behavior by @Sanderhoff-alt in https://github.com/vectorize-io/hindsight/pull/3038
• fix(coding-agents): stop the prompt hook wiping the retain cursor; write state atomically (supersedes [#3136]) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3412
• fix(knowledge): sync backing mental_model name on page rename (#3307) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3407
• fix(search): stop extreme relative date offsets from crashing recall (#3217) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3413
• feat(coding-agents)!: drop retainEveryTurns — batching belongs on the server by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3415
• fix(consolidation): cancel sibling tag groups when a batch fails by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3417
• feat(coding-agents): apply retain attribution to all ingestion paths by @voarsh2 in https://github.com/vectorize-io/hindsight/pull/3418
• fix(recall): fill the source_facts budget in rank order and flag truncation (#3221) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3419
• fix: untrack the coding-agents node_modules symlink by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3422
• fix(embed): define daemon stop() success by port occupancy, not the health probe by @Alan5168 in https://github.com/vectorize-io/hindsight/pull/3171
• feat(coding-agents): handle 429 on both the request and the poll path by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3423
• fix(consolidation): refresh mental models every round, not just the final one (#3411) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3416
• feat(coding-agents): optInOnly — run memory only in projects that were opted in by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3433
• fix(mental-models): trace delta-ops call and decouple its completion cap (#3421) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3424
• fix(graph-maintenance): drive the entity prune off a queue, not a bank-wide sweep (#3222) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3409
• fix(engine): scope document statements by bank_id (#3429) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3435
• retain: don't hold the data-plane connection across a separate-store write by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3414
• fix(coding-agents): bound the 429 retry by the caller's clock, not a constant by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3425
• retrieval: one unified store call for all arms + all fact types by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3434
• fix(files): give each batched file a unique storage key (#3226) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3437
• fix(llm): preserve inline <think> literals in _strip_reasoning_tags by @hundunweimi in https://github.com/vectorize-io/hindsight/pull/3426
• blog: Give Any Agent Plugins Client Long-Term Memory by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3440
• fix(reranker): batch FlashRank passages instead of one unbounded forward pass (#3355) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3441
• fix(release): ship Rust CLI license notices by @Sanderhoff-alt in https://github.com/vectorize-io/hindsight/pull/3447
• fix(api): correct stale OpenAPI examples by @Sanderhoff-alt in https://github.com/vectorize-io/hindsight/pull/3436
• fix(llamacpp): pass extra_body through factory to LlamaCppLLM by @handnewb in https://github.com/vectorize-io/hindsight/pull/3431
• docs(config): clarify that zeroing maintenance intervals stops scheduling, not execution by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3454
• fix(coding-agents): say when a turn is running without memory by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3455
• perf(recall): score observation expansion set-wise, not per candidate row (#3085) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3451
• fix(ci): fetch crates for all targets before the offline license scan by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3457
• perf(recall): make temporal extraction ~9x faster without changing behaviour by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3452
• fix(release): keep package-lock.json's version in step with the release by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3456
• refactor(kb): create pages and mental models atomically by @Sanderhoff-alt in https://github.com/vectorize-io/hindsight/pull/3369
• fix(coding-agents): run the opencode survey under our own agent, not plan mode by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3460
• feat(api): forward allowlisted request headers to extensions by @rares-mollie in https://github.com/vectorize-io/hindsight/pull/3428
• feat(docs): blog list infinite scroll, search, category badges, mobile polish by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3463
• recall: carry entity_ids on results, skip redundant re-fetch in entity_build by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3461
• feat(eliza): Hindsight long-term memory integration for elizaOS by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/2385
• fix(llm): send an explicitly configured reasoning_effort (#3449) by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3459
• docs(examples): add TEI embeddings + reranker docker-compose example by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3465
• Dedupe graph maintenance submits against running jobs, not just pending ones by @cdbartholomew in https://github.com/vectorize-io/hindsight/pull/3444
• blog: Claude Code Now Builds and Reads Its Own Knowledge Base by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3464
• docs: link Hindsight Academy from the docs navbar and footer by @benfrank241 in https://github.com/vectorize-io/hindsight/pull/3473
• feat(memories): assert_writable — let a store close a bank to writes by @nicoloboschi in https://github.com/vectorize-io/hindsight/pull/3489

New Contributors

• @JoshFunnell made their first contribution in https://github.com/vectorize-io/hindsight/pull/3310
• @dragonkid made their first contribution in https://github.com/vectorize-io/hindsight/pull/3289
• @Alan5168 made their first contribution in https://github.com/vectorize-io/hindsight/pull/3260
• @TheAngryPit made their first contribution in https://github.com/vectorize-io/hindsight/pull/3342
• @ishfuseini made their first contribution in https://github.com/vectorize-io/hindsight/pull/3398
• @hundunweimi made their first contribution in https://github.com/vectorize-io/hindsight/pull/3426
• @rares-mollie made their first contribution in https://github.com/vectorize-io/hindsight/pull/3428

Full Changelog: https://github.com/vectorize-io/hindsight/compare/v0.9.0...v0.9.1