Your memory, across all AI tools.
Work across every tool you use, all connected to the same memory, kept in one place.
Nothing gets lost along the way and you don't have to repeat yourself. Quick questions or your hardest problems — you or your team already have the context.
Works with the AI tools you already use
Connects over MCP, REST API, or CLI — so it works with any AI tool that speaks any of them, for coding, research, writing, or anything else. New tools and models are supported automatically as the ecosystem grows.
The problem no one's actually solved: memory is only saved in the tool that created it.
Many AI tools let you decide what to save, but the memory stays locked to the tool that saved it. Switch models, start a new session, pick up on a different device — and it's gone. With Stompy, whatever you save goes with you: any model, any session, any device.
Portability runs deeper than the interface, too — the same memory is available whether you're working through it directly, an AI assistant is working with you, or an agent is acting on your behalf, regardless of which server or environment it's running on.
Stompy isn't a passive store. Once you decide something's worth keeping, it gets evaluated against everything you've already saved — the same underlying check, doing four different jobs:
Checked, not assumed.
Every save runs against what's already stored first. Nothing goes in blindly.
Flagged, not buried.
If something contradicts what you already told it, it gets surfaced — never silently overwritten.
Versioned, not vanished.
Every change stays on record. Nothing's lost without a trace.
Connected, not siloed.
Save something that relates to an earlier decision, and Stompy finds the link automatically — what depends on it, what it depends on, what's already there.
This is the delta evaluation architecture described in a peer-reviewed paper presented at ICMCIS 2026 (NATO's International Conference on Military Communication and Information Systems) — the mechanism deciding what's new, what's an update, what's redundant, and what conflicts, before anything is stored. Active in how it processes what you've already decided to keep. Never active in deciding what counts as memory in the first place.
Three steps. Two minutes.
Connect your AI
Pick whichever AI tool you use — Claude, ChatGPT, Gemini, Perplexity, or one of the coding tools like Cursor or Codex. Each one connects a little differently. We'll show you exactly what to do for the one you pick.
Connect your AI →Save something worth remembering
A decision, a preference, anything worth keeping. Stompy checks it against what's already there.
Open a different tool
It's already there. No re-explaining, no catching up.
Simple, Transparent Pricing
Choose the plan that fits your usage. All plans include core Stompy features.
Free
For personal use
Usage Limits
- Up to 500 saved items, lifetime
- Unlimited personal projects
- Semantic search
- Document ingestion
- Community support
Pro
For regular use
Billed annually ($60/year)
Usage Limits
- Up to 300 saved items/month
- Everything in Free
- Create and share one project (up to 5 people)
- Priority support
- Advanced analytics
- API access
Power
For heavy users
Billed annually ($240/year)
Usage Limits
- Up to 1,200 saved items/month
- Everything in Pro
- Bulk operations
- Priority queue
All plans include
We'll email you when it's ready. We won't email you about anything else.
Your data is safe with us. Here's how.
Isolated by design.
Every project lives in its own database schema — nothing crosses over, and it's enforced by tests on every change we ship, not just true when we checked last.
Your own key, scoped to one project.
Mint an API key for a project and it can only ever see that project — it can't touch another one, rotate itself, or even confirm another project exists.
Verified every session.
Every session has to prove it's really you before it can read or write anything.
Encrypted in transit.
Everything moving between you and Stompy is encrypted, no exceptions.
Real deletion.
Delete a project and everything in it, any time — gone for good. You see exactly what's on the line, contexts and sessions included, before you confirm.
Give your AI a memory.
One config line. Every session builds on the last.
"No memories yet. Which is fine. Totally fine. Stompy's just here. Waiting."
Built out of spite for context windows that forget everything.
We ran the benchmarks that could've proven us wrong — and published them anyway. Tech nerds, this one's for you.
retrieval recall
tasks completed with memory (Haiku)
Sonnet with memory vs. memoryless Opus, every metric
recall and search round-trips, ~90ms down to 2–4ms, fleet-wide
Doesn't help yet on small, new projects (under about 2,000 lines of code) or work that's finished in a single session.
Frequently asked questions
Everything you need to know about Stompy
About two minutes. Connect your first AI tool, and you're ready to save your first memory.
Yes. One account, one memory — connect as many AI tools as you like, and they all read from the same place.
Doesn't matter. Search finds what you meant, not just what you typed.
Anything worth remembering — a decision, a preference, background on a project, a rule you want followed. If it's worth not repeating, it's worth saving.
Stompy flags the contradiction instead of quietly keeping both versions.
It connects them. Save something related to a decision you made earlier, and Stompy finds the link on its own — no extra tagging, no manual filing.
Reading, searching, and exporting everything you've already saved always keeps working, no matter what — saving something new is the only thing that pauses.
Still being finalised: Confirmed for the free tier's cap only — Pro and Power monthly-cap behaviour isn't decided yet.
No. Setup is a short guide you follow once — clicking through a few settings, not writing anything. After that, everything happens inside the AI tools you already use, in plain conversation.
Up to you. Keep it personal, or invite your team into a project so everyone's working from the same context.
In your dashboard, laid out plainly — not as a database query.