Everything people ask

A living map of what you've built. It holds a feature graph for each project, which is the tree of what the system is rather than the list of what's currently in flight, and it attaches the status, the tickets, the docs and the decisions to each node.

It reads from Linear and GitHub, and your coding agents write into it directly.

Technical teams shipping several projects where a good share of the code is written by AI agents. Solo founders with a portfolio, engineering leads at small startups, and consultancies who have to report status to clients.

Linear tracks work items: things in flight that will eventually be done. Stemmory tracks the system those items are changing, including features you've decided on but haven't built, which have no ticket and therefore no home in Linear.

Stemmory reads from Linear rather than replacing it. You'll keep using both.

You already have a docs folder and it's probably out of date. It goes stale because keeping it current is a separate act of discipline from doing the work.

Stemmory parses those same markdown files and derives status from your tickets, so the map updates when the work updates instead of when someone remembers.

No. It maps features, not files or functions. What it stores is your documentation and ticket data, never your source code. See the data questions below.

A GitHub repository, and optionally Linear. Both take one OAuth click. You can start with GitHub alone. Status derivation is the part that wants Linear.

Two changes, both small. Feature descriptions live in docs/features/*.md in your repo, one file per node, with a few lines of frontmatter. And your agents need an API key to write decisions.

Tickets, commits and reviews all stay exactly where they are.

The parsed path is configurable per repository. It defaults to docs/features.

Connecting takes a few minutes. Getting a useful graph depends on how much of the project you describe in those markdown files. One file per feature, and the tree assembles itself from the parent field.

Stemmory reads the markdown files in the path you configure, plus metadata from Linear, and stores what they yield — titles, frontmatter and short excerpts — not your source code, which it doesn't store or transmit. One caveat, stated plainly: reading any repository file uses GitHub's contents-read permission, which technically can read other files too, so the guarantee we stand behind is about what we store. The privacy policy spells this out.

Postgres, on Supabase. Every table has row-level security, so a query can only return rows belonging to your own workspace. That's enforced by the database itself rather than by application code that could forget.

No. v1 is a read-only mirror by design. Connecting Stemmory can't alter, close or comment on anything in the tools you depend on.

Everyone invited to a workspace can read it — viewers included. Per-project scoping for viewers is specified and not built yet, so do not rely on it to keep one client from seeing another. Privileged actions like billing, members and API keys are restricted by role, and each one lands in an audit log admins can read.

Each agent key is distinct, so every write carries which agent made it and when.

The human handle and device name behind a write are stored either way. What's off by default is whether your teammates can SEE them, which is a workspace-level setting. That distinction matters: in an agency context the display becomes performance surveillance, but we're not going to pretend we don't hold the value.

Paddle. Paddle.com Market Ltd is our authorised reseller and the Merchant of Record for every order, which means they're the legal seller. They take the payment, issue the invoice, and calculate and remit any sales tax or VAT owed where you are.

Your card statement and your receipt will name Paddle.

Yes — 14 days on every plan, and it runs at Portfolio level so you try the best version rather than the cheapest one. A card is required up front and the first payment is taken when the trial ends; we email you 3 days before that happens.

One click on your subscription page inside the app. No email, no support ticket, no retention call.

Paid access runs to the end of the period you've already paid for. After that the workspace goes read-only rather than away: the graph, the decisions and the journal stay there and stay readable, and sync stops. You can undo a pending cancellation any time before it takes effect.

Both directions, from inside the app, once your first payment has been taken — plans cannot be switched mid-trial. Monthly to annual is prorated against the time you've already paid for, so you pay the difference rather than a full year on top of a month you already bought.

Annual to monthly credits the unused remainder against future invoices. Neither one needs a support ticket.

Yes. A full refund within 14 days of your first payment, no questions asked.

Renewals aren't covered, because you can cancel from inside the app at any point during the term. Paddle issues the refund, since they're the Merchant of Record, and it goes back to your original payment method.

Full details are in the Refund Policy.

For new subscriptions, yes, any time.

Not for you. Once you are subscribed you keep the price you signed up at, for as long as the subscription continues — a rise in the list price does not reach existing subscribers. What can change is what a plan contains, and that only ever takes effect at a renewal, with at least 30 days' notice so you can cancel first.

Your workspace and its graph stay where they are. You lose access to the product, not the record.

Prices are listed in USD. Paddle shows your local currency and any tax due at checkout, before you pay.

Being specific, because publishing it is the fair thing to do. There's no MCP server yet. Agents write to the graph through API keys today, and the MCP interface that would let them read it is specified but not built.

There's also no unified home or activity feed. You navigate by project.

Agent API keys exist for writing into the graph. A general-purpose read API doesn't.

Stemmory is built by one person, and is early. That's a real limitation, and it also means a well-argued request reaches the person who writes the code.

The product is sold by Adyrix Systems Private Limited through Paddle. Support goes to support@stemmory.com and is answered by the person who built it.

support@stemmory.com, answered by the person who built the product.