The short answer
Start with a copied project and check which content reaches the agent.
Check attachments, links, corrections and data destinations before expanding.Open your Gigabrain knowledge in Obsidian
We recommend Obsidian for reading your notes and exploring their links. It is optional; your agent still runs Gigabrain.
- Download Obsidian from its official website.
- After Gigabrain setup, choose Open folder as vault and select your knowledge folder.
- Open Graph view. Click a linked note to read it.
- In Excluded files, keep
Corpus/,_retrieval/andraw/out of the graph.
Use your local system folder. Community plugins and Obsidian Sync are not required.
Graph view shows note links. Gigabrain’s tools inspect source evidence and typed relationships.
Keep your vault as the starting point
Obsidian's local Markdown notes give a file-capable agent useful source material. You still need to check which notes are current, which backlinks represent verified relationships and what access an AI integration should have.
Obsidian documents how it stores vault files and configuration. Preserve your originals and start the agent workflow on a small copy. Your vault, plugins, synchronization and agent provider each have their own behavior.
Choose a representative project, not the whole vault
Copy a few notes from a project you understand well: a project overview, a meeting decision and a later correction. Keep links and attachments needed to interpret them. Exclude private journals, unrelated clients and credentials. Record which folder is the original and which is the test copy.
Before importing, write five questions whose answers you can check yourself. Include a missing fact and an outdated statement. A successful demonstration should show you which source was used, not just produce a familiar-sounding answer.
Do not ask an agent to reorganize your original vault as the first step. Make a working copy, review the proposed file changes and check that you can restore the originals.
Use the supported Gigabrain setup sequence
Follow the release's START-HERE instructions and setup guide. Gigabrain requires Python 3.10 or later; the tested agents and systems lists the agent and operating-system combinations we tested.
sh bootstrap.sh
"$(cat .gigabrain/python)" tools/gigabrain_doctor.py --preflight
"$(cat .gigabrain/python)" tools/gigabrain_doctor.pyRun these commands from your extracted Gigabrain folder. On Windows, follow the Git Bash setup instructions. Check the tested versions listed above and resolve any setup errors before importing your working copy.
Use the import and indexing steps supplied with your release. Keep the source path recognizable so a citation can lead back to the original note. Gigabrain's own knowledge structure and derived retrieval material have different roles from your original vault.
Test the answer and the path back to the note
- Ask for the original project commitment and require a source passage.
- Open that file yourself and verify that it supports the answer.
- Add the correction through the supported workflow and refresh derived retrieval data.
- Start a new session and ask which commitment now applies.
- Ask the missing-fact question. Expect an explicit gap, not a plausible invention.
The public memory test tests local retrieval with fictional notes. It does not test your notes, plugins, chat application or the correction steps above; check those separately.
Understand where data can travel
| Operation | Boundary to inspect |
|---|---|
| Note storage and local retrieval | Files and derived index on your computer |
| Hosted AI agent | Selected material shared with that provider |
| Optional cloud graph authoring | Selected source excerpts sent to the provider after you give permission |
| Sync, plugins and remote Git | Independent services and destinations you enable |
"Local files" does not mean every connected operation stays on the device. Review each integration's settings, source scope and retention terms before sharing sensitive work.
Check what a copied vault leaves out
A note can look complete in Obsidian while depending on content outside the text itself. An image may contain a figure that appears nowhere in the text. An embedded note may provide the context. A plugin may display a computed view that is not present in the underlying Markdown.
For the first project, inspect these cases yourself. Do not assume the agent sees the same rendered view you do. Record which content reached the import and which still needs a separate extraction or review step.
- Attachments: check that required PDFs and images are present and supported by the import workflow.
- Embeds and links: open a linked note and confirm whether its contents were imported or only its link text.
- Plugin views: compare the visible result with the source file before relying on computed information.
- Renamed notes: repeat a citation check after changing a filename or moving a folder.
- Sync copies: confirm which device has the latest source before refreshing retrieval.
Support varies by plugin and attachment format. Record any material the import skipped so you know what the agent can search.
The Obsidian pilot checklist records these dependencies and the questions used to check them. If the first answer is wrong, the checklist helps you distinguish a source-copy problem from retrieval or reasoning. You can expand the import after the small project is reliable enough for its intended use.
Decide what to add next
If this small project produces useful, verifiable answers, add one new source type and repeat the tests. If it fails, keep the input small while you find out whether the issue is import coverage, retrieval, an ambiguous note or agent behavior.
Gigabrain is a separate knowledge system, not an official Obsidian product. Start with System. Add the Loop Pack if you want the prepared automations. Check integration fit before purchasing and use the workbook to record your first run.
