AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
docs(release): two coherence fixes, no new claims
xerj-org0f2cbca45a4f
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 528e8f31a..34ae63701 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -208,9 +208,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0original run gave — there is no signal that anything is now recoverable.With `--no-graph` the file is re-sniffed and then has to be re-admitted, which- is decided by four gates in `classify_new` (`reconcile_plan.rs:279-323`): the+ is decided by four gates in `classify_new` (`reconcile_plan.rs:279-323`) — thefamily/group filter, `ensure_compatible`, a field-overlap threshold, and a- refusal when two frozen datasets match equally well. All three outcomes are+ refusal when two frozen datasets match equally well — which between them select+ one of three outcomes. All three arereachable — the file is recovered in place, or the run aborts naming it, or itstays junk — and which one you get depends on what the other files in thecorpus froze into the plan. This entry deliberately does not predict it: sixdiff --git a/docs/recipes/air-gapped-deployment.md b/docs/recipes/air-gapped-deployment.mdindex 1296104c6..538a61225 100644--- a/docs/recipes/air-gapped-deployment.md+++ b/docs/recipes/air-gapped-deployment.md@@ -148,10 +148,15 @@ this was measured:dropped session leaves the tree". Also false, and the interesting part is how:the measurement behind it signalled the wrapping `bash script.sh` processinstead of the forked subshell that actually owns the trap, so the trap was- never asked to run. Signal the subshell and it runs — `SIGTERM`, `SIGHUP` and- `SIGINT` all clean, verified in a real pty with the signal landing provably- mid-`tar`. Three of this page's wrong sentences now trace to a harness that- measured the wrong thing rather than to the shell behaving unexpectedly.+ never asked to run. Signal the subshell alone and it runs — `SIGTERM`,+ `SIGHUP` and `SIGINT` all clean, verified in a real pty with the signal+ landing provably mid-`tar`. The exception, and the reason the lead singles it+ out, is `SIGTERM` to the whole process group — how `timeout`, a CI runner and+ systemd deliver it — where `rm -rf` is killed alongside the shell and the tree+ survives. Three of this page's wrong sentences trace to a harness that measured+ the wrong thing rather than to the shell behaving unexpectedly; this last+ distinction is a fourth thing the harness could not see, because it never+ signalled the group.Hence the form above: the body is `"$STAGE"`, relative to the directory theblock already `cd`-ed into and never leaves, and the `trap` is installed