floorianmb22 downloadsResearch your vault with ChatGPT on desktop and mobile, billed to your own ChatGPT plan.
Research your Obsidian vault with ChatGPT – on your iPhone.
Ask questions, let the agent search and read your notes, and have it write new ones.
Paid for by your ChatGPT plan, not an API key.
Install · Quickstart · Features · Skills · Privacy · FAQ
![]() |
![]() |
| Ask on your phone. "Create a quick demo view with Prism", then "Make a 3D view". The agent writes the note and checks that Prism renders it. | The note it wrote, rendered by Prism: an interactive 3D chart you can orbit with a finger. |
Real screenshots from the Obsidian iOS app.
Your notes sync to your phone. Your thinking partner should come with them. Vault Scout puts a research agent inside the Obsidian app on iOS. It searches your vault, reads the notes that matter, follows links and backlinks, and answers with citations you can tap. When you ask it to, it writes notes too, and only after you approve each change.
It runs entirely inside the Obsidian app. There is no server in between and no API key, and you don't need a computer. You sign in with ChatGPT once, and requests count against your ChatGPT Plus or Pro plan through OpenAI's Sign in with ChatGPT for open-source apps.
Vault Scout is not in the community plugin directory yet. The steps below work once the first release is published.
With BRAT (works on iPhone and iPad)
floorianmb/vault-scout.Manually
Copy main.js, manifest.json and styles.css from the latest release into <vault>/.obsidian/plugins/vault-scout/. If your vault syncs (Obsidian Sync with Installed community plugins on, or iCloud), the plugin reaches your iPhone by itself. Restart Obsidian there and enable it.
Requirements: Obsidian 1.13 or later and a ChatGPT Plus or Pro plan. Free accounts can sign in, but cannot use plan usage.
sequenceDiagram
autonumber
participant You
participant VS as Vault Scout (in Obsidian)
participant Safari
participant OpenAI
You->>VS: Sign in with ChatGPT
VS->>Safari: Open sign-in page
Safari->>OpenAI: Sign in, approve plan usage
OpenAI-->>Safari: Redirect to 127.0.0.1 (page doesn't load)
You->>VS: Paste that address
VS->>OpenAI: Exchange the code (PKCE)
OpenAI-->>VS: Tokens, stored on this device only
You->>VS: Ask a question
VS->>VS: Search, read and follow links in your vault
VS->>OpenAI: Request with the notes it found
OpenAI-->>VS: Answer with linked notes
Why the copy-and-paste step? Sign in with ChatGPT sends the browser back to a small server on 127.0.0.1, and an Obsidian plugin can't run one on iOS. Copying the address carries the one-time code over by hand. The code is useless without a secret that never leaves the plugin (PKCE).
| Search | Full-text search over your notes, with folder and tag filters. Results include frontmatter title, description and type plus matching snippets. If no note contains every word, it falls back to the best partial matches. |
| Read and follow | Reads notes (long ones in chunks), lists folders, and follows outgoing links and backlinks. Several reads run in parallel. |
| Cite | Every note it relies on is linked in the answer. Tap a link to open the note. |
| Write, with approval | create_note, edit_note (replaces one exact passage) and append_to_note. You see a diff and tap Apply or Reject. Hidden folders and the Obsidian configuration are off limits. |
| Properties | get_properties and set_properties read and change frontmatter through Obsidian's own frontmatter API, without touching the note's text. null removes a property. You approve the before and after. |
| Rename and move | rename_file renames a note or moves it to another folder, after you approve it. Obsidian updates every link to it. The file extension stays the same. Nothing is ever deleted. |
| Ask you | When something is unclear, for example which of two notes you mean, the agent asks with ask_user, with suggested answers as buttons, instead of guessing. |
| Selection | Select text in a note and run Ask about the selection (command palette or the editor menu). The question is scoped to that passage, and edits stay inside it unless you ask for more. |
| Images | Tap the image button next to the input to attach up to four photos or screenshots. They are scaled down on the device before they are sent. |
| Check Prism blocks | After writing a ```viz block, it opens the note so Prism renders it, reads Prism's error log and fixes what broke. |
| Remember | Chats are notes in Vault Scout/Chats/. Open an earlier one from the history button and continue where you left off, also on another device. |
| Pick the model | The model menu lists the models your ChatGPT account can use. |
Skills are instructions the agent loads when a task needs them. They use the same SKILL.md format as skills for Claude Code and Codex, and live in your vault under Vault Scout/Skills/, so they sync to every device.
AGENTS.md or CLAUDE.md taking precedence.SKILL.md. Paste its link.vault-scout-context that load along with it.check_prism instead of Prism's command-line render.On a Mac you can also import the skills you already use with Claude Code and Codex.
api.openai.com under your ChatGPT account. OpenAI's terms for Sign in with ChatGPT apply. Requests are made with store: false.api.github.com, raw.githubusercontent.com) when you install or update a skill. There is no analytics or telemetry, and no server of ours.data.json.Does this cost API credits? No. Requests use the allowance of your ChatGPT Plus or Pro plan. There is no API key and no API billing.
Can it change or delete my notes behind my back? No. Every change, including a rename or move, is shown for approval first (Settings → Confirm changes). There is no tool for deleting notes.
Why do I sign in again on my iPad? Sign-ins are per device on purpose. OpenAI replaces the refresh token on every refresh, so a synced session would keep logging your devices out of each other.
Do I need a Mac or a server? No. Everything runs in the Obsidian app on your iPhone or iPad.
dynamic_agent_client), plus a per-device ext_agent_host_id. ID token claims (iss, aud, nonce, sub) are checked. The token comes straight from the token endpoint over TLS (OIDC Core 3.1.3.7).stream: true and store: false, as plan usage requires. On mobile the event stream is read in one piece through Obsidian's requestUrl. The desktop app streams through Node's https module, since the web view blocks fetch for these responses.store: false, every request resends the conversation. Reasoning carries over as encrypted_content. Older tool results are shortened so long chats stay within the context window.eslint-plugin-obsidianmd recommended config, with no warnings.npm install
npm run build # type check and bundle main.js
npm run lint # official Obsidian lint rules
npm run deploy -- <vault> # copy the build into <vault>/.obsidian/plugins/vault-scout
skills/ contains the skills offered under Install skills, installed from this repository.
Several features are ideas from Chatting with AI by o1xhack, itself based on obsidian-chat by Omar Shahine (both MIT). Thank you! Chatting with AI is a mobile-friendly vault agent that also supports Anthropic and OpenAI API keys. Have a look at it.
Adapted from it: keeping sign-in credentials in Obsidian's SecretStorage, the frontmatter tools get_properties and set_properties, ask_user, scoping a question to the editor selection, image attachments, and rename_file. Vault Scout implements them on its own and adds approval to every change. No code was copied.
MIT © floorianmb