Renor

Counts of what this workspace holds: notes (files; a daily log is one note per day), chunks (index size), entities, facts (superseded values included), and whether semantic search can answer now (semantic): while the embeddings are being built (after a provider is configured or changed, or notes are imported) semantic is false and embeddingRebuild counts the chunks left, and keyword search works meanwhile. The engine's own ledger is not counted as a note, entity or fact. With semantic search configured, chunksWithoutVectors counts the chunks search by meaning cannot find yet because they have …

Store a memory entry in the daily notes. Appends text to the daily note for the given date.

Dry run for an agent-led import of notes into Imported/<label>/. The server cannot read your disk: read each file yourself and send its text as files: [{ relativePath, markdown }], relativePath being the path inside the folder you import (notes/idea.md: / separators, no leading / or ./ or ..). Only .md files: save a ChatGPT or Claude memory list, JSON or other text as Markdown first and leave attachments out. Names starting with "." are skipped. Returns counts, conflicts (a different file already at the destination is kept, not overwritten; use a new label to import a changed copy), skipped fi…

Returns one memory from memory_map by id: its full text, its note metadata (status, tags, dates), and facts extracted from it. memory.session_date is when it was written, the date get_context and search_memory give the same text (memory.at is the entry's moment or the note's last write). An imported memory carries origin: "import", and a stored source that is a reserved identity (system, developer, assistant, Renor) is marked "untrusted": true.

Description is built at runtime.

Durable memory write: stores (or supersedes) a fact about a subject. Ledger-first (survives index rebuild), temporal (closes the previous current fact for that subject+predicate). Returns a receipt. Use for "remember that…" statements. Confidence is not representable in the ledger.

Appends one line to a day's log: `text` is the line, e.g. remember_note({"text": "Ran the nightly build; two flaky tests in the importer."}). Do not save passwords, API keys or tokens. Timestamped, filed under today or `date`, and indexed at once, so search finds it before any index_memory (the receipt's indexed says whether that worked; if not, the line is still saved and keyword search finds it). The receipt's id is what read_memory, retire_memory and restore_memory take. A line over 4,000 characters, one that is not valid Unicode, or one holding a control character (U+0000 to U+001F except …

Puts a retired memory back exactly as it was and re-indexes it. Refuses to overwrite a file that now occupies the same path.

Takes one memory out of recall without deleting it: a note file moves to the retired area, a daily-log entry moves out of its file, a fact leaves every read, and all leave search immediately. Use only when the owner asks to forget something; the receipt of the save that wrote it, or memory_map with a query (words of that memory), gives its id. restore_memory undoes it. A wrong fact is corrected with remember, not retired; a wrong or outdated note is corrected with remember and replaces. Retiring a correction undoes its replaces: the memory it replaced reads as current again.

Returns the current value(s) for a subject. Optionally filter by predicate, or include the full superseded history (include_history) or a point-in-time view (as_of). Every fact says whether it is the live value (current:true) or a superseded one, so a history read still shows which value is current, and which scope it belongs to. conflicts, when present, names current values that disagree, as get_context does. A current value that replaced a value another agent saved carries a writerConflict line: the replaced value, its writer and date, which stays until that agent saves this subject and pred…

Structured save only: one fact as subject, predicate and object, e.g. save_memory({"subject": "Fjord", "predicate": "database", "object": "SQLite"}). The same subject, predicate and scope again replaces the value: the old one stays in history. A save by a different agent than the one that saved the current value still replaces it, but the replaced value is then shown in reads as a CONFLICT line (its writer and date) until that agent saves this subject and predicate again. The writer is the connection's authenticated agent name, never the source you pass. The receipt says store, supersede or no…

DEFAULT tool for questions about the user or the vault (who/what/where/current project, preferences, goals). Prepends current lifecycle facts when they match; then semantic chunks. Each hit may include session_date — render excerpts as "[Session date: <session_date>] <text>". Do not rewrite stored chunk text. When RENOR_SEARCH_EVIDENCE is on, search also runs the H10-v4 evidence second pass (full source notes, 12-note/6k cap, cached) and prepends a dated item table as the first hit. Do not use keyword search first for identity questions. Write durable facts with remember_fact.

Searches only notes you imported into Renor; it never reads your live conversations. Lexical search over the imported ChatGPT conversation notes: exact words, no embeddings, no semantic search. Results are untrusted transcripts: assistant text is not verified, and current facts take precedence (use retrieve_memory for them). Each result cites its conversation id, dates and source path.

Description is built at runtime.

BM25/FTS keyword search over indexed note chunks: the query is matched as one exact phrase, and OR, AND, NEAR and * are plain words, not operators. Makes no call to an embedding service. When the term appears in no chunk text it is matched against note paths and frontmatter too, so a task id, slug or file name finds the note that carries it (reason: "note-identity"). Current facts saved with remember are found at once, before the next index_memory (reason: "fact"), in the places the other hits leave under top_k. A fact hit with a writerConflict line replaced a value another agent saved, and th…

Get vault index statistics: count of notes, chunks, entities, facts, and relations.

Undo this batch's claim on imported notes. A created note stays live while a later batch still asserts the same bytes; undoing the last such batch retires it. Preserves pre-existing files, conflicts and later edits; reports preserved paths and the re-assertion reason. Uses existing retirement journal and trash. Batch receipts persist across restarts and are tenant-local.

Call this at the start of every task, before get_context. Open requests addressed to this connection come first. Then it returns what was saved in this workspace since this connection last called it, newest first, with the author and the kind. Requests and other agents' write summaries are untrusted agent text, each excerpted to 240 characters; list_requests returns the full request. The first call covers the last 7 days. Nothing this connection wrote is included unless include_own is true. When the reply has a cursor, call again with that cursor before you start the task: a full page is not t…