Skip to content

armgit / xerj

Public Imported from GitHub
0

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.

docs(rerank): publish xerj hub sync in "Every way data leaves a XERJ node"

xerj-org27ed81e2f236
diff --git a/docs/RERANK.md b/docs/RERANK.md
index 170c8913e..b0b179c02 100644
--- a/docs/RERANK.md
+++ b/docs/RERANK.md
@@ -362,6 +362,7 @@ client appears in a source file this list does not account for:
| **`xerj autoindex s3:// --watch`** (client, not the node) — `--interval`, `--endpoint-url`, `AWS_*` | The same S3 requests as above, repeated every interval: a `ListObjectsV2` page per cycle plus a GET per changed object. Billed to the operator's object-store account, so the run keeps a spend ledger and stops at its budget | Every `--interval` while the watch runs | Off: only with `--watch` on an `s3://` root |
| **`xerj corpus build`** (client, not the node) — recipe `[[sources]]` `url` | HTTPS GETs to the source URLs your recipe names (one per `http-zip` source per build, plus the `git clone`/`fetch` a `git` source runs). Records travel INTO the machine; no document text and no credentials are sent upstream — it is a read-only fetch | While `xerj corpus build` runs | Off: only when a recipe declares an `http-zip` or `git` source |
| **`xerj hub pull`** (client, not the node) — `XERJ_HUB_URL` | GETs to the hub (`https://hub.xerj.org` unless `XERJ_HUB_URL` names another): the signed corpus catalogue (a conditional GET that carries the ETag of your cached copy, when there is one) and its `.sig`, then the one pack you named and its `.sig`. The requests name the catalogue and that pack, so the hub can see which corpus you pulled. Packs travel INTO the machine; no document text, no query text and no credentials are sent upstream — it is a read-only fetch. The pack is then restored into the node URL you gave the client | While `xerj hub pull` runs; a catalogue cached within the last 24 hours is reused without a request | Off: only when you run `xerj hub pull` |
+| **`xerj hub sync`** (client, not the node) — `XERJ_HUB_URL`, `--install-timer` | The signed corpus catalogue and its `.sig` from the hub, as for `xerj hub pull` but on every run (a conditional GET that carries the ETag of your cached copy, when there is one), not reused for 24 hours. For each pack you already hold whose pin moved, `xerj hub sync` then runs `xerj hub pull`: the new pack and its `.sig`, from the `pack_url` the signed catalogue names (a GitHub release asset in the published catalogue). The requests name the catalogue and the packs you update, so the hub and GitHub can see which corpora you hold. Packs travel INTO the machine; no document text, no query text and no credentials are sent upstream — it is a read-only fetch | While `xerj hub sync` runs: once when you run it, and every 6 hours for as long as the systemd user timer or launchd agent that `xerj hub sync --install-timer` writes is enabled | Off: only when you run `xerj hub sync`, or enable the timer it wrote (the command writes the file and enables nothing) |
A node started with the defaults opens none of these connections, and makes no
telemetry, update-check or licence call. Two things outside the node complete