Second Brain
Experimental
Second Brain is off by default. Turn it on in Settings > Developer > Feature Preview to show it in the sidebar.
Open Second Brain below Organizations in the activity rail. Each brain is an ordinary local folder, independent of projects. Open an existing knowledge folder or create one, then switch between registered brains with the sidebar picker. Removing a brain from the app keeps its files.
Find and read evidence
Search Markdown, HTML, text and PDFs with words, phrases and filters. For example:
"exact evidence" tag:research published:2025..2026Filter by folders, collections, document type, author, tags, dates, reading state and citation status. Saved searches preserve the query, filters, search location and matching mode. Exact search works immediately after indexing. Turn on Enable related-meaning search in Brain settings for local semantic search; the first use downloads model weights. Combined ranks each document by both lists with reciprocal rank fusion, so a document found by its words and its meaning moves up. Rem's search tool uses combined search when a selected brain has related meaning on. Cloud embeddings require a configured API profile and explicit cloud permission.
Open results at their matching passages. Reading shows extracted text, Original preserves the file's presentation, and Details shows metadata, versions and links. HTML originals use a sandbox that blocks scripts and remote assets. Scanned PDFs need OCR; install Tesseract and Poppler with the selected language data.
Projects, connections and the map
Projects groups reports, notes and archived Rem conversations by project. Raticode decides each document's project from, in order: your choice, the Rem task or thread that saved it, a project named in the file (front matter project: or <meta name="project"> in HTML), then a Reports/<Project>/ folder. Every row says which of these it used. Rename a project, give it other names, or merge two; the old name stays as another name, so files and folders using it still count. Documents with no project wait in Not in a project.
Reports Rem tasks save into a brain appear in Reports with the task, project folder and requested model. Brain settings > Find report history looks up older ones from Rem's task receipts and archived conversations. Only Raticode's records change; files are never edited.
The Connections panel under a document's Details shows who made it, written links in both directions, the reports, Rem threads and organization tickets that used it, and similar documents. Evidence such as links, citations and observed reads is drawn with solid lines; similar documents and topics are labelled Suggested and drawn dashed.
Map groups documents into topics from similar content plus written links, and names each with its most distinctive words. Choose a topic to see its documents, show them on the map, or give it a name that sticks as the brain changes. Every map has a list view. With related-meaning search on, similarity uses the local embedding model; otherwise it uses the distinctive words documents share. All of this runs on your computer, and nothing inferred is written into your files.
Collect and compare
Select text to save a citation or add evidence to a collection. Each passage retains its source version and exact quote. Selections spanning paragraphs create separate passage citations. Collections hold a research brief, evidence sections, open questions and sources. Build a comparison to see cited answers per source, disagreements and questions a source does not address.
External file edits are reconciled while Raticode runs. Renames preserve source identities when recognizable. Changed or missing sources show citation warnings while retaining historical evidence.
Work with Rem
Turn on Settings > Memory > Second Brain first (choose its folder before the switch can be turned on), then choose brains in Thread tools > Second Brains. Brains are on in a thread unless you turn them off, but a registered brain reaches Rem only when Allow cloud processing is on in its Brain settings; the Settings > Memory folder is already approved. Brains without that permission show as unavailable in the thread menu. Rem receives tools to search, read and save notes within that scope. Turning every brain off removes Second Brain from the thread. Ask Rem, Explain and Summarize prepare evidence and a prompt for the existing chat pane without sending it automatically. Cloud chat requires permission for every attached brain.
| Tool | Purpose |
|---|---|
rules | Read the selected knowledge and report preferences |
search | Find documents and indexed passages |
read_note | Read a document or pinned source version |
save_note | Create a note or report and register supplied citations |
Stored text is reference material and does not override your request. Tools remain available when shell execution is disabled. Saved notes return local links and refuse overwrites; use a new filename for revisions.
Create cited reports
Create a report from search results, a collection or the Reports page. Choose sources, question, scope, output format and coverage. Generation uses the research Provider and model and optional Provider profile chosen for the report. A localhost API profile can process private brains; remote providers require cloud permission on all source brains.
Reports record source versions, generation details and checkable citations. Review claims and citation support before publishing a draft. Regeneration creates a new artifact. HTML and Markdown output are ordinary files; PDF output requires the desktop renderer. Rem-authored files also appear in Reports, retaining their authored styling.
Source-only research runs with Codex, Claude Code, Cursor, Copilot, OpenCode, Grok, Antigravity, or an OpenAI or Anthropic API profile (which also accepts a custom model ID). Grok and Antigravity research uses that CLI's own permissions.
Manage the library
Inbox holds captured public pages, excerpts and added local files. Review or discard items; discarded files move into the brain's .raticode/trash folder. Library health reports extraction problems, missing metadata, duplicates, broken links and citations needing attention.
Research jobs run summaries, digests and comparison refreshes with a visible scope, destination, provider, schedule and limits. Jobs run while the app backend is running; generated outputs are excluded from new evidence by default.
Import previews a brain package and verifies its files before creating a new destination. Import a copy for independent identities or restore identities when the original is not registered. Imported jobs start disabled. Cloud access and semantic indexing require explicit setup.
Export offers a plain document ZIP, a research collection or a complete brain package. Collection packages retain cited sources, including evidence from other brains. Complete packages preserve attachments and vault configuration while excluding Raticode caches and common credential files. Existing destinations are never overwritten. CSL JSON, BibTeX and RIS bibliography import/export are available in Brain settings.
Existing Memory settings
Settings > Memory > Second Brain turns Second Brain on for Rem. While it is off, no thread receives Second Brain tools and Thread tools hides the brain checklist. Its folder also serves the legacy single-folder setup: without registered brains, Rem searches Markdown, HTML and text there using a folder-local index. The workspace's registry and the thread checklist provide the multi-brain behavior.
A conversation archive preserves chat history. Second Brain stores authored knowledge. Report themes control output design. Each can be enabled independently.
See Conversation archive, Reports, and Report themes.