The Chronicle

The Chronicle is a weekly review written by The Librarian itself. It gathers what changed, what the collection learned, and what remains open, then files the result as a normal Markdown reference. It gives you a durable answer to “what happened here last week?” without depending on a chat transcript or a model’s memory.

Chronicle is off by default. Configure it under Settings → Curator → Chronicle.

What it writes

Each completed week gets an entry at:

references/chronicle/YYYY-Www.md

For example, ISO week 31 of 2026 is references/chronicle/2026-W31.md. Because the entry is a reference, agents can find it with search_references, and you can read, edit, or restore it from the Vault. Re-running the same week updates that week’s entry instead of creating a duplicate.

Every entry has a deterministic digest built from local evidence:

• vault changes committed during the period;
• memories created, updated, proposed, archived, or otherwise changed;
• handoffs created or claimed, plus questions that remain open;
• Intake and Grooming outcomes when those runs can be attributed safely; and
• warnings where the available data cannot support a trustworthy claim.

The evidence appendix is written even when no language model is configured. Intake and Grooming runs record their shelf attribution, and Intake records the actual prompt and completion usage reported by its model calls. Older run records that predate those fields are labelled unavailable instead of being guessed.

Optional narrative

Choose a provider and model in the Chronicle tab if you want a readable narrative and up to three possible blog seeds above the factual digest. This step is optional and fail-soft: an unavailable provider, malformed response, or narration error still produces the deterministic entry and records the run as Digest only.

Before facts are sent to the configured provider, secret-shaped values are redacted across the complete payload and the prompt is bounded. The provider token itself is never placed in the prompt or logs.

Schedule and Run now

The default schedule is Monday at 08:00 in the server’s local time, but it does nothing until you enable it. A scheduled run reviews the previous completed week. If the server misses the chosen time, it catches up on the next scheduler poll rather than skipping the week.

Run Chronicle now is deliberately different: it writes a partial review of the current week, even while the automatic schedule is disabled. The entry and run history are labelled partial so they cannot be mistaken for a completed week.

Shelves and privacy boundaries

Chronicle writes one entry per writable system shelf. It never mixes facts from different shelves into one review. On the standard single-shelf installation this simply means one entry in your vault.

On a multi-shelf installation, Chronicle includes only the Intake and Grooming runs attributed to the shelf being reviewed. Legacy records without an explicit shelf belong to the main shelf; they are never copied into another shelf’s review. Read-only shelves are skipped.

Reading the run history

The Chronicle tab records the ISO week, shelf, trigger, completion status, narration mode, model, token usage, entry path, and a value-free failure label. The server log contains only the same operational counts and outcome metadata — never memory text, handoff text, prompts, or the generated narrative.