Watch on YouTube: https://www.youtube.com/watch?v=grEM3BDBnfo
Channel: Kacper Rutkiewicz | AI Made Simple
Giving GrokBot a Persistent Memory with Obsidian
Most AI assistants treat every conversation as a blank slate. You feed them context, they answer, and the moment the session ends the knowledge evaporates. Kacper Rutkiewicz’s latest demo shows a practical alternative: connect GrokBot to a local Obsidian vault so the bot reads and writes plain‑text files that you own, version, and can inspect at any time. The result is a shared “second brain” that survives across agents, sessions, and even machines — without locking your data into a proprietary vector store.
What the Video Covers
- Why a folder of Markdown files beats opaque memory APIs for business facts.
- End‑to‑end setup: creating the vault, installing GrokBot’s native Obsidian connector, and configuring a dedicated VaultBot agent that curates the vault.
- How to give every other bot a single read‑only instruction so they all reference the same source of truth.
- Security boundaries, cost transparency, and a live ingest of a meeting transcript.
Step‑by‑Step Walkthrough
1. Create the Vault
Make a local folder (e.g., Grockbot Vault) and open it in Obsidian. Inside, create five top‑level folders: sources, wiki, raw, inbox, and trash. This structure keeps raw inputs separate from curated pages and gives the bot a predictable place to drop new notes.
2. Install the Native Connector
In GrokBot, open Marketplace → Connectors, search “Obsidian,” and install the official connector. There is no vault picker at install time; you will name the vault later in the bot’s instructions.
3. Add the VaultBot Curator Agent
Create a new agent called VaultBot and paste the provided curator prompt. Its rules:
- Show a plan before any write operation.
- Never delete — move obsolete files to
trash. - Cite source documents for every claim.
- Never store secrets, passwords, or client‑confidential data.
- Run a weekly cleanup check every Monday morning (broken links, drift from sources, self‑corrections).
4. Share a Single Read‑Only Instruction
Every other bot on the account gets one line in its system prompt:
Read wiki index first. Only read needed pages. Never edit wiki/raw. Save new notes in inbox with YYYY-MM-DD_botname_topic.
VaultBot periodically files those inbox notes into the proper wiki pages.
5. Lock Bots to This Vault
By default, any bot on the account can reach every Obsidian vault on the machine. Add a rule to each bot: “Only use Grockbot Vault. Never read or write any other Obsidian vault.” Reads do not ask permission; writes do.
6. Ingest Source Documents
Drop price lists, FAQs, SOPs, meeting transcripts, or PDFs into sources. Tell VaultBot to ingest. In the demo, a team meeting transcript produced a new “drain camera inspections” wiki page, updated pricing and FAQ entries, and refreshed the index — all for roughly 1 % of the weekly $30 plan allowance (total demo ≈ 6 %).
7. Run and Schedule Cleanup
Trigger a manual cleanup once to verify the pipeline, then enable the automatic Monday‑morning routine.
8. Grab the Free Resources
The video description links to a free resource guide and the exact VaultBot prompt. Enter your email to join the “AI Made Simple” School community and download them.
Tools, Numbers, and Costs
- GrokBot plan: $30/week (usage meter shown in UI).
- Demo usage: 4 % → 5 % after one ingest; full demo ≈ 6 % of weekly allowance.
- Obsidian: Free (Obsidian Sync is optional paid).
- VaultBot cleanup: Runs every Monday morning automatically.
- Security test: 31 unsolicited connection attempts to email/Slack/calendar over 5 days — all blocked.
Caveats and Hype Check
- No revenue promises. The method itself is presented as a productivity pattern, not a money‑making scheme.
- Lead magnet. The free guide and prompt are gated behind an email signup to the creator’s School community.
- Cost transparency. The demo shows ~6 % of a $30/week plan for a small vault; larger vaults will cost more, and the creator explicitly says he cannot guarantee costs at scale.
- Security honesty. Reads do not prompt for permission; the video demonstrates this and advises keeping client files, journals, and passwords out of the vault.
- Platform dependency. Obsidian must be open and the computer on for the bot to reach the vault (at least on first run; subsequent runs may work with Obsidian closed).
- Personal branding. A closing prayer segment is part of the creator’s style, not a product pitch.
Concrete Next Actions
- Create a local folder (e.g., “Grockbot Vault”) and open it as a vault in Obsidian.
- In GrokBot, go to Marketplace → Connectors → search “Obsidian” → install the native connector.
- Add a VaultBot agent with the provided curator prompt (plan‑before‑write, no‑delete, cite‑sources, no‑secrets, weekly cleanup).
- Add the shared read‑only instruction to every other bot: “Read wiki index first. Only read needed pages. Never edit wiki/raw. Save new notes in inbox with YYYY-MM-DD_botname_topic.”
- Lock each bot to this vault: “Only use Grockbot Vault. Never read or write any other Obsidian vault.”
- Drop source documents (price lists, FAQs, SOPs, meeting transcripts) into the vault’s
sourcesfolder and tell VaultBot to ingest. - Run a manual cleanup check once, then enable the Monday‑morning routine.
- Download the free resource guide and VaultBot prompt via the description link (email signup to “AI Made Simple” School community).
Final Thought
Don’t make it remember the answer. Make it remember where the answer lives.
By treating a plain‑text Obsidian vault as the canonical knowledge store, you gain auditability, portability, and a single source of truth that any agent — current or future — can read. The setup is lightweight, the ongoing cost is measurable, and you remain in full control of the files. If your workflow already leans on Markdown notes, this pattern is a low‑friction way to give your AI assistants a durable, shared memory.