Search every AI coding
conversation on your Mac
Claude Code, Cursor, Codex, Zed, Kiro and Antigravity each save your chat history to disk, in their own format, where you can’t search it. Inventory reads all six and turns them into one index you can search by keyword — or by what you half-remember. ⌘⇧Space from any app. Nothing is uploaded.
No account, and no card to try it. One-time $19.99 after that, for lifetime access.
You already solved it. That bug, that config, that command. It is sitting in a chat you had three weeks ago. Inventory finds it before you finish typing.
Every agent you use.
One place to look.
Inventory reads the local stores your tools already write to, read-only and snapshotted first, and merges them into a single searchable timeline. Search your Cursor chat history, dig up an old Claude Code session, or find that Zed conversation from last month — one search box covers all of it.
Cursor
Claude Code
Zed
Codex
Kiro
AntigravityFour steps, then never think about it again.
Install the desktop app
One download for macOS. No account, no onboarding call.
It finds your tools
Inventory locates the local stores Cursor, Claude Code, Zed, Codex, Kiro and Antigravity already write to.
It indexes in the background
First pass takes seconds. After that it stays live as you work, reading each file once.
Hit ⌘⇧Space anywhere
Full-text and semantic results across every tool before you finish typing.
The last search bar your history will ever need.
Search by meaning, not just keywords
Every conversation is embedded on-device and blended with keyword results, so “container stuck” finds “pod stuck terminating”. Hits found by meaning are labelled, because a result with none of your words otherwise looks like a bug.
Your agents can search it too
Switch on Agent access and Claude Code, Cursor and any MCP client can search your history mid-session, finding what you already solved instead of asking again. Read-only, this Mac only, and every call they make is listed for you to see.
Quick capture that knows your history
⌘⇧N saves a thought and immediately matches it against everything you've already worked out. It surfaces the conversation from last week where you solved it.
Resume where you left off
Claude Code and Codex sessions reopen in the terminal with the full transcript attached, so the agent picks up with real context, even if the project folder has moved.
Hand off to any tool
Copy a condensed primer of any thread with what was being solved, the last exchanges and the code, then paste it as the opening message wherever you like.
Clipboard scratchpad
⌘⇧V recalls anything you've copied, tagged with the app it came from. Off by default, and honestly labelled about what it stores when on.
You choose how much history
Index the last 7, 30, 90 or 365 days, or everything. The command palette shows the on-disk size of each choice, so the trade is visible before you make it.
Ranked by what worked, not by what matches.
Every other history search ranks by similarity, which scores the answer you abandoned exactly like the one that fixed it. Inventory reads the outcome instead — from what you said next, and from whether the error came back. Flip the switch.
- 1Bump the CI test timeout to stop the flakedidn’t work
Raise the Jest timeout to 20s — the suite is probably just slow on CI's shared runners, not actually broken.
Cursor•3 weeks ago•the same test failed intermittently two runs later - 2Notes on flaky CI testsno verdict
Flaky tests are usually shared mutable state, a timing assumption, or unseeded randomness colliding between parallel runs.
Codex•5 months ago•no outcome either way — nothing was tried - 3Tests fail in CI only, pass locally every timeworked
It was Date.now() — two tests were racing on the same second boundary. Freezing the clock in test setup fixed it.
Claude Code•6 days ago•you replied “yep that was it, zero flakes since” - 4Flaky suite after adding parallel test workersworked
Two test files were writing to the same fixture at once. Giving each worker its own temp directory stopped the collisions.
Zed•2 months ago•the failures stopped showing up for the rest of the week
The top hit is the answer you already tried and that didn’t work. It reads as the best match because it is one — similarity has nothing to tell it apart from the fix with.
Conversations shown are illustrative. The research note has the method and the real numbers, measured against 500 conversations.
Everything you copy, one keystroke away.
Hit ⌘⇧V and a running list of everything you've copied slides open, tagged with the app it came from. Off by default. On the moment you want it.
Ask it the way you remember it.
You won't remember the exact words. You never do. Type what the problem felt like and Inventory still finds the conversation, because it matches by meaning as well as by keyword.
Mostly about the part we worried about.
The local-only decision is the interesting one. I've got two years of Cursor and Claude Code history sitting in places I can't search, and the reason I've never tried a cloud tool for it is that those conversations contain client context I'm not willing to upload.
keychain-held key plus opaque file on disk is exactly the right shape for this, a stolen backup or synced drive won't hand someone your whole tool history. appreciate you actually shipping it instead of just noting it for later
It is a really nice product and it has much more opportunities further. We can use this indexed data for multiple use cases.
cross tool search for ai chats is such a usefull idea 🙌 context is so scattered rn
The things people ask before buying.
What does Inventory actually do?
Does any of my conversation history leave my Mac?
Which tools does it read?
Is this a memory layer? Does it feed context back into my AI tool?
How is this different from my tool's own history, or from grep?
What does it cost?
Does it work on Windows or Linux?
More detail in the docs, on security, or in the comparison against built-in history and cloud AI memory.
Ready to index your mind?
Free for 7 days, then $19.99 once. No subscription, no account.
Payments secured by Polar