Repository navigation
fix(api): consolidation discovery probe, mental-model edit race, LlamaParse v2 - #5513
Merged
Merged
Conversation
…aParse v2 - banks_needing_consolidation(): per-bank EXISTS probe instead of JOIN + GROUP BY over memory_units, which seq-scanned the table on large backlogs (#5425) - update_mental_model: re-check name/content under the row lock and re-embed if a concurrent edit moved them, so the stored vector matches the stored text (#5297) - LlamaParse: port the connector to the v2 parse API; v1 returns 410 for v2-only accounts (#5447)
…ote raw embedding read in race test
nicoloboschi
force-pushed
the
fix/api-consolidation-mm-llamaparse
branch
from
October 12, 2026 09:53
4b115b0 to
bbbc41c
Compare
This was referenced Oct 13, 2026
Closed
Closed
nicoloboschi
added a commit
that referenced
this pull request
Oct 13, 2026
…e launch video main reverted #5513, which removes its two changelog entries and the 6ab9509eeb56 migration — the release now ships no new migrations. Also embeds the launch video (placeholder id) below the TL;DR, reusing the iframe pattern from the 40k-stars post.
nicoloboschi
added a commit
that referenced
this pull request
Oct 13, 2026
…pshot (#5550) * docs(blog): Hindsight Memlake launch post (draft feature branch) - blog/2026-10-08-memlake.md: 'Introducing Hindsight Memlake: serving 100M Memories at <100ms p99 on object storage', with figures, charts and cover - featured: pin the Memlake post, unpin the 0.10.2 release post - announcement bar: the post title, linking to the post - custom.css: .post-wide for the post's wide figures * docs(blog): Memlake post dated 2026-10-13 (slug, featured pin and announcement link updated) * docs(blog): Memlake is written in Rust * docs(blog): simpler Rust paragraph * docs(blog): Rust in one sentence * docs(blog): reasons for Rust * docs(blog): smoother Rust sentence * docs(blog): add the dedicated cloud instance deployment option * docs: Cloud & Enterprise section (overview, Hindsight Cloud, Enterprise with Memlake) * docs: changelog and version-1.0 docs snapshot for v1.0.0 - Generated changelog entry for 1.0.0 (v0.10.3..main). - Snapshot docs/ as versioned_docs/version-1.0/ and register 1.0 in versions.json, so 1.0 becomes the served docs version. - Document HINDSIGHT_API_BANK_ALIAS_CACHE_TTL_SECONDS and _MAX_ENTRIES, which the memory-banks page already linked to but configuration.mdx never defined (dead anchor in the build). - Regenerate the hindsight-docs agent skill. No package/SDK version bumps: release.sh does those when the tag is cut. * docs(blog): wire the Memlake whitepaper link and the 49k star count Whitepaper publishes at https://vectorize.io/memlake/whitepaper; the 'architecture, guarantees and tradeoffs' link now points at the Enterprise page's Memlake section, which this branch adds. Official star count is 49k. * docs(blog): clear the Memlake draft blockers and link the quickstart All open questions resolved, so the review comments come out. The quickstart placeholder now points at /developer/api/quickstart. * docs: regenerate the 1.0.0 changelog after the #5513 revert, embed the launch video main reverted #5513, which removes its two changelog entries and the 6ab9509eeb56 migration — the release now ships no new migrations. Also embeds the launch video (placeholder id) below the TL;DR, reusing the iframe pattern from the 40k-stars post. * docs(blog): swap in the real launch video * docs: regenerate the 1.0.0 changelog against the released tag v1.0.0 is cut, so the entry is now generated from the real tag range (v0.10.3..v1.0.0, 19 commits). Still no new migrations. The tag-scoped rejection is filed under Breaking Changes: it rejects calls that used to return bank-wide results. Also syncs the docs skill's openapi.json to 1.0.0. * docs(figures): narrate the Memlake architecture diagram scripts/docs_skill_figures.py --check-all requires every figure to render as text for the docs skill, and the architecture diagram shipped with no scenes. Adds one scene walking the API, the servers, the bucket and the background jobs.
This was referenced Oct 14, 2026
nicoloboschi
added a commit
that referenced
this pull request
Oct 15, 2026
…r matches the stored text (#5619) A document edit embedded from a pre-image read off the lock; a concurrent edit could move the half the caller did not send, and the write then stored a vector for text the model no longer holds (#5297). The edit now takes the row lock, re-embeds only if that half moved, and takes the history snapshot from the locked row. Re-lands the #5513 part reverted in 65ef1bd. Fixes #5297
This was referenced Oct 16, 2026
nicoloboschi
added a commit
that referenced
this pull request
Oct 16, 2026
* fix(api): port the LlamaParse connector to the v2 parse API (#5447) v1 (/api/parsing) returns HTTP 410 for v2-only LlamaCloud accounts, so file retain with the llama_parse parser failed before conversion. v2: POST /api/v2/parse/upload with a configuration (tier + version), poll GET /api/v2/parse/{id} on job.status, fetch expand=markdown_full. Recreates the LlamaParse part of #5513, which was reverted together with unrelated changes in 65ef1bd. * fix(api): review fixes — parse the LlamaParse upload reply into a model, note why v2 next to the URL
This was referenced Oct 16, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three core bug fixes in hindsight-api-slim.
6ab9509eeb56reinstallsbanks_needing_consolidation()as a per-bankEXISTSprobe instead ofJOIN memory_units ... GROUP BY. Same results, no full table scan on big backlogs. Wrapper (lock_timeout, skip arms, install gating,execute_unless_owned) copied unchanged. PostgreSQL-only routine, so the Oracle slot is left out on purpose.update_mental_modelnow takes the row lock for document edits too, re-checks name/content under it, and re-embeds (and re-takes the history snapshot) if a concurrent edit changed them. Common path still embeds off the lock.POST /api/v2/parse/uploadwith aconfigurationfield, pollGET /api/v2/parse/{id}onjob.status, readmarkdown_fullviaexpand. Output is still one markdown string. v2 requires a tier, so it sends{"tier": "cost_effective", "version": "latest"}(closest to v1's default cost). No new config.Each fix has a regression test that fails without it.