pensar

LLM wiki engine for R. Your knowledge base grows while you work.

An open source R implementation of the Obsidian markdown vault concept combined with Karpathy’s LLM wiki pattern. The LLM maintains the wiki. You edit, curate sources, and ask questions.

pensar (“to think”) has one dependency: yaml.

Install

remotes::install_github("cornball-ai/pensar")

The idea

Most people’s experience with LLMs and documents is stateless. You ask a question, the LLM retrieves some chunks, generates an answer, and forgets everything. Nothing compounds.

pensar takes a different approach. An LLM reads your sources (articles, chat logs, project briefings, whatever you point it at), writes wiki pages that synthesize the key ideas, links everything together with [[wikilinks]], and maintains an index. The knowledge base gets richer with every source you add and every question you ask.

The vault is the synthesis layer, not a data lake. Sources that already live somewhere (corteza sessions, saber briefs) get referenced, not copied. Content without a home (a link someone sent you, a pasted article, a quick note) can be ingested directly.

Quick start

library(pensar)

# Create a vault (pick a path you actually want to look at)
init_vault("~/wiki")
use_vault("~/wiki")   # persist for the session; add to ~/.Rprofile

# Ingest a source
ingest("Article content here...",
       type = "articles",
       source = "https://example.com/interesting-post",
       title = "Interesting Post")

# Check what's in the vault
status()

# Find what links to a page
backlinks("Interesting Post")

Configuring the vault path

Per CRAN policy, pensar never writes to a default home-filespace location. You have to point it somewhere. Resolution order (first hit wins):

1. The vault = argument (or path = for init_vault()) when passed explicitly.
2. The PENSAR_VAULT environment variable.
3. Walk-up from getwd() for a directory containing schema.md (project-local vaults).
4. options("pensar.vault"), set by use_vault() – typically in ~/.Rprofile as a global default.

If none of these resolves, pensar errors with a setup hint. There is no implicit fallback. Same shape for vault_export()’s out_dir: pass it explicitly or set PENSAR_SITE_DIR.

Vault structure

{vault}/
  raw/
    articles/       clipped articles, pasted text, links worth preserving
    chats/          conversation logs worth keeping
    briefings/      project briefings (one per project, historical record)
    matrix/         messages from Matrix rooms
  wiki/             LLM-maintained pages (summaries, concepts, analyses)
  index.md          auto-generated catalog of everything
  log.md            append-only record of operations
  schema.md         conventions for content in the vault
  CLAUDE.md         instructions for Claude Code when started here
  AGENTS.md         same content for Codex and other agents
  {name}.Rproj      RStudio project file

raw/ is for content you want to preserve in the vault. ingest() stores it there. Sources that already live somewhere and don’t need preservation can be referenced directly by wiki pages in their frontmatter, no ingest needed. index.md and log.md are maintained by pensar functions.

Working with an AI agent

init_vault() seeds CLAUDE.md and AGENTS.md by default so any agent you start in the vault (Claude Code, Codex, etc.) knows how to operate on it — what files are immutable, how to drill down with pensar show, when to rebuild the site, and so on.

For conversational use, start your agent session in the vault directory itself. The working directory becomes the knowledge base, auto-memory stays scoped to vault work, and file edits land in the right place by default.

Pass agent_instructions = FALSE to init_vault() if you don’t want these files.

Versioning: git or syncthing?

Use both, for different things:

• Git for the vault source (raw/, wiki/, index.md, log.md, schema.md, etc.). The vault is plain markdown — it diffs beautifully, history matters when a wiki page gets revised, and you can push to a private GitHub repo for backup. After init_vault(), just run git init && git add . && git commit -m "initial vault".
• Syncthing (or Dropbox, etc.) for the rendered site (vault_export() output), so you can browse on your phone without running anything. Set PENSAR_SITE_DIR to a synced folder and pensar export writes there. (vault_export() requires either PENSAR_SITE_DIR or an explicit out_dir = – there’s no implicit cache fallback.)

Don’t sync the vault source via Syncthing. Concurrent edits from multiple devices on the same .md file get messy, and you lose history. Use git for that.

Note: “raw” in pensar terminology means source documents in raw/ (vs. synthesized wiki/ pages), not “raw text”. Everything in the vault is markdown — there’s no separate raw-vs-rendered distinction inside the vault itself. Rendering happens via vault_export() into a separate directory.

Functions

A pensar CLI is also installed at {pkg}/bin/pensar:

pensar status              page counts by category
pensar lint                health check
pensar show "<page>"       drill-down inspection
pensar back "<page>"       backlinks only
pensar tag <tag>           pages with this tag
pensar log [n]             last n log entries
pensar export [out-dir]    render to static HTML

Symlink {pkg}/bin/pensar to somewhere on your PATH (e.g., ~/.local/bin/pensar) to use it as a command.

Conventions

Every page uses YAML frontmatter and plain [[wikilinks]]. Obsidian-style aliases such as [[page-slug|display text]] are supported: pensar resolves page-slug as the target and renders display text for readers. View the vault in RStudio, Obsidian, any markdown editor.

---
title: Page Title
type: concept
source: https://example.com
date: 2026-04-13
tags:
  - R
  - testing
---

This connects to [[Other Page]] and builds on [[Raw Source]].

Sister packages

License

Apache 2.0