Skip to content

v0.13.2

Choose a tag to compare

@github-actions github-actions released this 27 Sep 13:09
· 7 commits to main since this release
chore(release): 0.13.2 on Cognitive Nexus 0.14.2

Dependencies (Cargo.toml):
- anda_cognitive_nexus 0.14.2 (from 0.14.0), with the 0.14.1 query work.
  0.14.2 (anda-db d781b47) fixes connect, whose KIP 1.x layout probe left a
  handle without setup that Store::open adopted. Under 0.14.0 every Brain
  Space, reopened after creation, ran its Concept collection without the
  Nexus index hooks and with the default tokenizer; under 0.14.1 new
  Concepts also missed `query_keys` (a promoted draft type queried by its
  installed name returned nothing) and a Space 0.14.0 wrote failed to load.
  The first load under 0.14.2 upgrades the collection schemas and builds
  `query_keys`/`lookup_key` over existing rows; the changelog carries the
  upgrade notes and the not-rebuilt BM25 limit.
- The DB/KIP/Core requirements are `0.14` / `0.16`; the lockfile resolves
  anda_kip 0.14.0 and anda_core/anda_engine 0.16.2, which Brain needs.
  scripts/sync-kip-reference.mjs and the Worker's codegen-prompts.mjs read
  the resolved anda_kip from Cargo.lock instead of an exact Cargo.toml pin,
  which the requirement no longer has.
- tiktoken-rs 0.12.1 within 0.12.x, and mimalloc 0.1.52 (v3 is the default
  since 0.1.49; the `v3` feature is gone).

Recall tokenizer identity: `o200k_base@tiktoken-rs-0.12` (was `…-0.12.0`).
The old identity is the same encoding and stays accepted from requests,
Memory Interface budgets and stored MemoryPolicy.recall_budget values; it is
read as the new name, and resolved budgets, packets, receipts, the
descriptor and the recall_memory tool schema report the new one. The Worker
advertises the new name and accepts both. The semantic Watch evaluator pin
digests the identity, so a semantic binding needs reconfigure_evaluator and
re-arming, as for any evaluator change.

The handler test that raced the scheduled maintenance a Space load kicks
now waits for the Space to be idle.

Versions: anda_brain, the CLI and the skill report 0.13.2; the Worker
package is 0.1.2. Docs (API, README, semantic Watch runtime, SKILL copies,
AGENTS.md, Worker README) follow.

Checks: cargo fmt --check; cargo clippy -p anda_brain --all-targets
--all-features -D warnings; RUST_MIN_STACK=16777216 cargo test -p
anda_brain --all-features on Nexus 0.14.2 (570 + 17 + 11 + 13 + 4 + 3,
including the two tests 0.14.1 failed) and --lib (376); CI=true pnpm
--filter @ldclabs/anda-brain-worker check (codegen, tsc, 136 tests, dry
run); sync-kip-reference --check --worker and sync-kip-assets --check
against anda_kip 0.14.0; anda-cli go vet and go test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>