01
Install the app
Grab the macOS, Windows, or Linux build from the downloads page. On first launch the vault is created at
~/.1aivault/vault.db — a single SQLite file you own.Fig. 00 — Docs
A guided tour of 1AIVault.
Six screens, three jobs: pull every AI conversation into one vault, classify it locally, and serve it back to every tool via MCP. Here's how each piece works.
Fig. 01 — Quick start
Up and running in three steps.
No accounts, no cloud, no JSON editing. Download the app, point it at your AI tools, and your context is portable.
02
Import your conversations
Open Collect from the sidebar. 1AIVault auto-detects Claude Code, Codex CLI, Cursor, Cline, and Claude Desktop on your machine. Hit Import All or toggle Watching per tool to keep new conversations flowing in.
03
Install the MCP server
Go to Connection → Install MCP. One click writes the config for every detected AI client. Restart the client and it can now read from your vault during chats.
Local-first — nothing is uploaded by default
Optional Ollama for offline topic classification
Fig. 02 — Dashboard
Your vault at a glance.
The first view every time you open the app. Counts, classification progress, source coverage, and what your AIs did while you were away.
Counts strip
Entries, decisions, preferences, topics, connections, skills, and sources — updated live as new conversations flow in. The green delta (+10) shows what landed since you last opened the app.
Classification meter
Shows how many entries have been classified into topics. Hit Classify Now to run a batch — local Ollama by default, or any model you point it at.
Usage histogram
Entries bucketed by how often AI tools have read them: Unused, Used (1–4×), and Frequent (5+×). The buckets are clickable — they filter the classify view.
Source pills
Which AI tools have written to the vault. A green All synced badge means every detected tool is up to date.
Recent activity
A reverse-chronological feed of saves with the source pill, the snippet, and a category tag. Click a row to jump straight to the entry.
Tip
The dashboard is empty on first launch. The empty state walks you through importing your first conversations and installing the MCP server.
Fig. 03 — Collect
Pull conversations from every AI tool you use.
The Collect view is the funnel. It scans for AI tools on your machine, surfaces what it found, and lets you import on demand or watch for new conversations automatically.
On machine
Detects local AI tools by looking for their config files and conversation stores. Shows the count of conversations found and a status pill — Auto-importing, Ready, or Detected.
Watching vs Import
Watching means new conversations are pulled automatically as they appear. Import is a one-shot pull. Toggle Watching per tool from the same row.
Browser Extension
For ChatGPT and Claude on the web, install the companion extension. It mirrors completed conversations into the vault with the same source attribution and topic classification.
Other
Catch-all for imports that don't fit a tool: pasted conversations, JSON exports, and Obsidian-style markdown folders.
Rescan / Import All
Rescan re-detects tools after you install a new one. Import All pulls everything detected across the page in one batch.
View entries
Each tool row has a View entries → link that jumps to the classify view pre-filtered to just that source.
Fig. 04 — Topics
Topics discovered, not configured.
A local LLM groups entries by topic across every source. Each topic gets a digest, an alias list, source attribution, and a timeline.
Digest
A short summary the classifier writes for each topic — what you've been working on, written in a tone you can scan. Toggle Markdown and Raw from the header.
Aliases
Synonymous topic names that get merged into the canonical one. Add your own to fold related topics together.
Sources
Which AI tools contributed entries to this topic. A topic with Claude Code + Cursor sources is one you tackled across multiple agents.
Timeline
Every entry in the topic, oldest at the top, with the source pill and a snippet. Click a row to open the full conversation.
Reclassify
Re-run classification just for this topic with the current model and threshold. Useful after editing prompts or pointing at a better local model.
Classify Now
The bottom-of-page button kicks a batch over the pending entries. It shows a progress strip while running and respects your local rate limits.
Fig. 05 — Graph
See your knowledge cluster.
The same topics, drawn as a force-directed graph. Bigger nodes are denser topics. Edges are shared entries — places where two topics legitimately overlap.
What the edges mean
Shared entries: the same conversation contributes to both topics. Entry links: explicit links drawn between entries. AI declared: the classifier said two topics are related.
AutoLinker
A background job that proposes new links between topics based on semantic overlap. Accept, reject, or ignore from the right panel.
List ↔ Graph
Use the toggle in the Topics header to flip between flat list and graph. The selection persists — pick a topic in the list, hit Graph, and that topic is highlighted.
What's this?
The legend has a built-in explainer. Hover any line type to see a one-sentence definition of what it represents.
Fig. 06 — Connection
Install MCP once, read everywhere.
The Connection view is mission control for the bundled MCP server. Install it into every supported AI client, then watch activity in real time.
Install MCP
One-click install for Claude Desktop, Cursor, Claude Code, Cline, and Windsurf. Writes the right config in the right place — and is idempotent, so re-running is safe.
Activity
A live table of every MCP tool call: time, client, tool name, classifier, and the matching detail. Filter by event type or classifier and tick Auto-refresh.
Memory Reads
A focused slice of Activity: just the read-side calls (vault_search, vault_brain_query, vault_recent), with the query, result count, and latency.
Diagnostics
Server status, transport (stdio / HTTP), database path, WAL state, and a button to copy a support bundle if something misbehaves.
Verify a connection
After installing into a client, restart that client and ask “what's in my vault?” or “what have I been working on this week?”. If the answer is grounded in your real work, you're connected. You'll see the matching vault_search call appear under Activity in 1AIVault within a second.
Fig. 07 — Memory Reads
See what your AI actually read.
Every search your AI runs against the vault shows up here with the query, the matching entries, and how long it took. This is how you know the AI is using your memory — and what it's missing.
Filters
Filter by client (Claude Desktop, Cursor, …), tool (vault_search, vault_brain_query, vault_recent), and a time window (last hour, last 7 days, all time).
Result cards
Each card shows the query, the entries returned (color-coded by source), and the latency. Click an entry to open it in the vault.
Top contributors
The right rail surfaces the entries that get read most often. Those are the high-signal ones worth pinning so they never age out.
No memories returned
When a search returns nothing the card is flagged in coral with a “possible vault gap”note — these are the searches that show you what your AI wants to know that isn't saved yet.
Read often, never updated
A side panel that flags entries that are read frequently but haven't been edited in a while. Good candidates for a refresh.
Tip
If you're wondering whether your AI “remembers” the conversation you had yesterday, look here first. If there's no read for it, it never had a chance.
Fig. 08 — Data
Move your vault to a new machine.
The vault is one SQLite file. You can move it manually, or use the built-in cross-device transfer to ship it through any channel as a passphrase-encrypted bundle.
Vault location
~/.1aivault/vault.db on macOS and Linux, %APPDATA%\1AIVault\vault.db on Windows. Open Vault Folder jumps you to it in your file manager.Export vault…
Bundles the whole DB plus skills and configs into a single passphrase-encrypted file. Pick a passphrase and a destination — that's the whole flow. Send it via AirDrop, email, or a USB stick.
Import from file…
Pick the bundle on the new machine, enter the passphrase, and 1AIVault unpacks it into the local vault path. Existing entries are merged by ID, so re-imports are safe.
Backups
Daily snapshots live under
~/.1aivault/backups/ for the last seven days. You can restore one manually by replacing vault.db when the app is closed.Fig. 09 — Shortcuts
Keyboard first.
A handful of shortcuts cover most navigation. Most of what you need is one chord away.
| Shortcut | Action |
|---|---|
| ⌘ K | Open the command palette / search vault from anywhere |
| ⌘ N | New manual entry |
| ⌥ Space | Global capture hotkey — save a thought from any app |
| ⌘ 1 – 4 | Jump between Dashboard, Collect, Topics, and Connection |
| ⌘ , | Open Settings |
On Windows and Linux, use Ctrl instead of ⌘.
Fig. 10 — Troubleshooting
When something doesn't click.
Most issues are surfaced inside the app — but here are the common ones and where to look.
No conversations detected in Collect
The tool needs to have written at least one conversation since its last reset. Hit Rescan and confirm the path under View entries matches where that tool actually stores history.
MCP install failed
Open Connection → Diagnostics. If the client isn't detected, install the client first, then come back and re-run the install. Re-running is safe.
AI doesn't seem to read the vault
Open Memory Reads. If there are no reads, the MCP server isn't registered with that client yet — restart the client. If there are reads but empty results, the AI is searching for things you haven't imported.
Classification is stuck
Classify Nowruns against a local model (Ollama by default). The dashboard banner shows whether Ollama is running. If you'd rather use a remote model, point it at one under Settings → General.
Vault feels slow
The first import populates the FTS5 index — give it a minute on very large histories. Steady-state search runs in well under 50 ms even at 10k+ entries.
Where do I file a bug?
Open Diagnostics and hit Copy support bundle, then email it to [email protected]. The bundle has no vault content — just app version, OS, and logs.
Ready to install?
Bring every AI conversation under one roof.
Free to download. Pro is a one-time $29 — no subscription, no cloud lock-in, your vault stays on your machine.