Deeplake, in plain language
The apiary keeps its memory on Deeplake, the database built for AI by Activeloop. It is the part of the stack that holds what your assistants learn. Here is what it is, why it is unusual, and why it means you own your data and pay almost nothing to keep it.
DownloadWhat it actually is
Most databases are good at one kind of question. A spreadsheet-style database is great at exact lookups: find the note that mentions this exact word. A vector database is great at meaning-based search: find the note that is about this idea, even in different words. Deeplake does both in one place, so the apiary can find the right memory by the words you used and by what you meant, in a single query. It stores plain text, structured fields, and the meaning-vectors side by side, as columns in the same table.
Read about securityHow your memory is stored and recalled
- 01
A note becomes a row
When the apiary distills something worth keeping, it writes a row into Deeplake: a short headline, a longer summary, the full original, and a meaning-vector that captures what the note is about. Nothing is thrown away, and every row carries where it came from.
- 02
The row keeps its history
Deeplake keeps a full version history instead of overwriting in place. When a fact changes, the apiary adds the new version and marks the old one replaced. You can always see what was true before, and nothing is ever silently lost.
- 03
Recall searches two ways at once
When your assistant needs context, the apiary searches Deeplake by exact words and by meaning together, then hands back the few notes that fit. Because both kinds of search run against the same store, there is no second system to keep in sync.
- 04
It lives where you choose
Deeplake is serverless and can run in your own cloud account, so the rows that hold your memory sit in storage you control. You are not renting a black box that holds your data hostage.
Why a store built for AI matters
A plain database was designed for rows of numbers and text, in a world before AI. It can be bolted onto meaning-based search, but the vectors live in a separate system and the two drift apart. Deeplake was designed the other way around: vectors and structured data are first-class together, so the store the apiary writes to is the same store it searches, with no seam in between.
This is also why memory in the apiary can be inspected and repaired instead of trapped behind someone else's closed recall api. Every note is a real row you can read, scope, and version. When a fact goes stale, the newer version supersedes it and the history stays. The memory gets sharper as it grows rather than turning into a pile you cannot audit.
Deeplake grew out of Hivemind, the open-source agent-memory project from Activeloop that the apiary builds on. The practical promise is simple: your memory is yours, it is cheap to keep, and it is durable. Deeplake gives the memories somewhere honest to live, and the apiary gives every assistant one consistent way to use them.
Deeplake is the database built for AI by Activeloop: a store that holds exact data and meaning-vectors together, keeps full version history, and can run in your own cloud. The facts on this page are rewritten in the apiary's own words from Activeloop's public materials.
Common questions
Do I need to understand databases to use the apiary?
No. You never touch Deeplake directly. The apiary reads and writes it for you, and the hive dashboard shows what has been remembered. This page exists for the curious, not as a requirement.
What does you own your data mean here?
Deeplake is serverless and can run in your own cloud account on Google Cloud, Azure, or Amazon. The rows holding your memory sit in a bucket you control, inside your own perimeter, rather than on a vendor server you cannot see.
Why is keeping memory so cheap?
Deeplake runs analytics-style queries directly against low-cost object storage, which Activeloop reports as far cheaper than traditional cloud data platforms at scale. Keeping notes around does not run up a big bill.
Is anything ever lost when memory changes?
No. Deeplake keeps a full version history. When a fact changes, the new version is added and the old one is marked replaced, so you can always see what was true before. Nothing is overwritten silently.
Keep your memory on a store you own.
Install the apiary with one command and give your agents a memory that lives on Deeplake, in a store you control.
Windows (PowerShell): irm https://get.theapiary.sh/install.ps1 | iex
Download