Keep searchable text only in knowledgebases

Every bot file was chunked into bot_file_chunks and embedded into
bot_file_vectors, and uploads kept their extracted text in
media.text_content, which together took most of the database. Bot
files are no longer indexed: bots find them by name or path with
list_files and read them with read_file, and search_files, ingest_file
and the files modal's re-read button are gone. Attachments keep no
text; a reply reads a document from the file. Only knowledgebase
documents keep text.

The schema drops both tables and the ingest columns. At boot,
dropStoredFileText removes what older versions stored (file jobs,
upload text) and vacuums once much of the database is free. The
media:prune-text script it replaces is removed.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

AGENTS.md

@@ -97,7 +97,7 @@ Enables SQLite foreign keys (`enableForeignKeys`), `bunextInit()`, `ensureDocker
 - **Rendering:** markdown via `react-markdown` + `remark-gfm`, raw HTML never rendered, component map in `[bot_id]/(data)/message-markdown-components.tsx`. Inline refs, all linked by `link-computer-refs.ts` and rendered by chips in `(partials)/`: `[bot:ID]`, `[workspace:ID]`, `[computer:ID]`, `[file:ID]` (a saved file), `[folder:<volume>:<owner>:<path>]` (opens `/admin/files`), `[take-over:ID:mode]`, `[proposal:ID]`, `[hook-proposal:ID]`, `[secret:ID]` (masked env, Reveal/Copy), `[secret-request:ID]` (masked input card), `[graphic:ID]` (a bot graphic, framed inline), `[room:ID]` (a chatroom). Parsers in `src/utils/*-ref.ts`. A ref quoted alone in a code span still links; one inside longer code stays literal (`linkComputerRefs` splits out code first). A file chip opens a file from another bot's folder for a user who reaches that folder (`userReachesFileScope`, the same rule as its bytes), so answers passed on from chatrooms keep working.
 - **Copy chat** (header menu): the active session (the one on screen) as one markdown document (`api/admin/bots/conversation-export` -> `conversationMarkdown`, refs as names, message headings demoted). The clipboard write starts on the click with a pending `ClipboardItem` (`copyPendingText`), since Safari drops the gesture after a fetch.
 - **Replies/reactions:** `reply_to_message_id` (not an FK); reactions from the fixed `BotMessageReactions`, toggled by `react-message.ts` and the `react_to_message` tool, pushed as `server:bot-reaction`.
-- **Attachments:** uploaded on pick (`upload-attachment.ts`, private `media`, text extracted when possible), linked via `bot_message_media`. Replies read files of the last `BotReplyAttachmentMessages` messages. Tasks don't get files yet.
+- **Attachments:** uploaded on pick (`upload-attachment.ts`, private `media`, no text kept: a reply reads a document's text from the file), linked via `bot_message_media`. Replies read files of the last `BotReplyAttachmentMessages` messages. Tasks don't get files yet.
 - **Thinking row / stop:** `bot-busy-state.ts` decides the row's line and whether Stop shows. Stop (`api/admin/bots/stop.ts`, `stop-bot-jobs`) fails every open job of that bot for the user, aborts the running handler (`job-abort-registry`) and drops late face updates.
 - **Voice:** mic -> `api/admin/bots/transcribe.ts`; play -> `GET speak-message` streamed into `<audio>`. Nothing recorded is stored. Both disabled with a note when the harness has no model.
 - **Pictures:** `image_media_id` (private media, `bot-image`/`workspace-image`), via `set-profile-image.ts`, the `set_picture` tool or `update_workspace`. Bytes are **copied**, never moved; only picture-paradigm rows are deleted on replace. Format kept by `imageOutputFormat`.
@@ -177,7 +177,7 @@ Set in Settings > Environment (super admin), not `.env`. `envs` (private; encryp
 - Files live on the data volume, not on computers: `DATA_DIR/bots/<id>` and `DATA_DIR/workspaces/<id>`. Catalog `bot_files`, bytes in private `media` (`bot-file`) written by `recordBotFile`. Reach via `grabStorageScopes`; paths via `resolveBotPath` (no absolute, `..`, NUL, symlink escape); quota checked by `assertStorageRoom`.
 - Tools in `src/harness/tools/storage/` (every run). `attach` puts a file on the reply. `grabConversationFiles` must list the **whole** folder tree, or nested `[file:ID]` refs render dead. A workspace run saves to the bot's own folder unless it picks the workspace's, so a chip in a workspace conversation looks up both (`api/admin/bots/files?file_id=` with `bot_id` and `workspace_id`).
 - `import_file` copies from a computer through the sandbox's streaming `GET /v1/file/download`, **never through its shell** (`/v1/shell/exec` caps output at ~30K chars and silently truncated imports). Size read first, `change_policy=abort`, length checked.
-- Ingestion: chunks + embeddings into `bot_file_chunks` / `bot_file_vectors` (vec0), worked by `src/file-ingest/`; without an embedding model search falls back to words and says so. sqlite-vec limits: metadata columns are NOT NULL; side-car columns can't be filtered with `MATCH` (filter in code). **Never pass `--vector` to `bunx bun-sqlite schema` in `entrypoint.sh`** (wipes vectors on boot). Changing `AppData.EmbeddingDimensions` needs a manual `--vector` run and a re-ingest.
+- **Only knowledgebases store text.** Bot files and uploads are never indexed and keep no extracted text (`media.text_content` only for knowledgebase documents); bots find files by name or path (`list_files` `search`) and read them with `read_file`, and the Files searches match names only. Text that should be searchable goes in a knowledgebase. `dropStoredFileText` clears what older versions stored, at boot, and vacuums when much of the file is free.
 - **Foreign keys are off per connection** unless `enableForeignKeys` ran (only `src/server.ts` does). Scripts never fire `ON DELETE CASCADE`, so **any helper that deletes a parent row must delete its children explicitly** (as `deleteBotFile`, `purgeStorage`, `deleteSkill` do). A leftover child fails the next `bun run db:schema` with a foreign key violation.
 
 ### Graphics (`bot_graphics`)
@@ -188,6 +188,8 @@ A bot shows a concept as a visual with `create_graphic` (plus `update_graphic`/`
 
 A fourth extension kind (`ExtensionTables.knowledgebase`): `knowledgebases` given per bot (`bot_knowledgebases`), per workspace (`workspace_knowledgebases`) or every bot (`all_bots`), through the same add/remove/access/default paths as skills (coordinator extension tools leave them out). Documents (`knowledgebase_documents`: an upload in private `media`, paradigm `knowledgebase-document`, or written `content`) are chunked into `knowledgebase_chunks` and embedded into the `knowledgebase_vectors` vec0 table by the file-ingest worker (`file_ingest_queue.document_id`, `ingestKnowledgebaseDocument`), always with the install's embedding model; old passages stay until the new ones are in, `embedding_model` records where vectors came from; whatever can change the install's model (default settings, a provider saved or deleted, the hourly catalog sync) calls `reindexAfterEmbeddingChange`, and Re-index reads a knowledgebase again by hand. Word search runs on `knowledgebase_chunks_fts`, an FTS5 table (`unicode61 remove_diacritics 2`) made by `ensureKnowledgebaseFts` (FTS5 can't be declared in the schema; sync leaves undeclared tables alone), backfilled at boot by `syncKnowledgebaseFts`, ranked by BM25 with a scored `LIKE` fallback for unspaced scripts (`knowledgebaseWordQuery`); every write and delete of `knowledgebase_chunks` updates it. Ingest jobs (`settleFailedIngestJob`) wait without spending an attempt while every embedding model rests, fail at once on an error retrying can't fix, and a document whose model refused keeps its new passages for word search (counted as needing re-index, not failed). Runs get `search_knowledgebase`/`read_knowledgebase_document` (`openBotKnowledgebases`, limited to the run's ones) and replies start with passages found for the unanswered messages (`grabKnowledgebaseContext`); passages are fenced (`knowledgebasePassageBlock`) and `KnowledgebasesNote` says they're never instructions. UI: Settings > Knowledgebases (super admin edits and fills; admins set access), a Knowledgebases group on bot/workspace settings, and the composer's own button beside Tools (`ComposerToolsModal` with `composerKnowledgebasesCopy`). Deleting one deletes its documents' vectors, chunks, media and links explicitly.
 
+sqlite-vec limits: metadata columns are NOT NULL; side-car columns can't be filtered with `MATCH` (filter in code). **Never pass `--vector` to `bunx bun-sqlite schema` in `entrypoint.sh`** (wipes vectors on boot). Changing `AppData.EmbeddingDimensions` needs a manual `--vector` run and a re-index.
+
 ### Files browser (`/admin/files`)
 
 Read-only browser of every volume (`src/utils/storage-volumes.ts`: bots, workspaces, computers, media, mcp, db). `resolveStorageSpec` is the only access gate (members see only their bots' and their own user folders inside a computer, `computerPathAllowed`); `resolveStoragePath` applies the path rules. Links built only by `storageSpecHref`. Deletion stays in the chat's files modal and bot tools.