Definition: EVE Memory is the persistent, project-backed context layer behind Taskade EVE, the meta-agent powering Taskade Genesis. Instead of hiding context in a black-box vector database, Taskade EVE saves what it learns as real Taskade Projects in a projects/memory folder. You can read it, edit it, and share it across your workspace.
TL;DR: EVE Memory has two scopes. Workspace memory lives in
projects/memoryand stays with the workspace. Account preferences live in one short file that follows you across workspaces and that you edit under Settings → Account.
Why EVE Memory Matters in 2026
Most AI agents store memory in opaque embedding stores you cannot read or edit. Taskade EVE flips that pattern. Every memory it writes is a Taskade Project, so you can open it, share it with collaborators, and audit what the agent knows. This is the Workspace DNA loop in action: Execution creates Memory, Memory feeds Intelligence, Intelligence triggers Execution.
Readable memory also protects against stale facts. An agent that repeats an outdated note as fact spreads the error into every later answer. When the note is a project you can open, you fix it in one edit. See Context Rot for why long-lived context degrades without care.
How EVE Memory Works
- Capture. During Taskade Genesis app generation, Taskade EVE notes structural decisions, naming conventions, and lessons from errors.
- Write. EVE saves the notes as projects in
projects/memory. Older workspaces can still show aprojects/memoriesfolder. EVE reads both names, so older memory keeps working. - Recall. On the next prompt in the same workspace, EVE reads the relevant memory before it plans, so the new build respects prior choices.
- Edit. You can open any memory project, remove outdated context, or add constraints by hand. The agent reads what you wrote.
The Two Scopes
| Scope | Where it lives | What it holds | Who edits it |
|---|---|---|---|
| Workspace memory | projects/memory in the workspace |
Project decisions, a running ledger, saved knowledge | You and EVE |
| Account preferences | /user/PREFERENCES.md |
How you like to work, across every workspace | You and EVE, or Settings → Account |
Workspace memory stays with the workspace, so collaborators share the same context. Account preferences belong to you. They follow you into every workspace and stay separate from workspace memory. You can read and change them under Settings → Account.
The Task Ledger
Each space keeps a TASKS.md ledger inside projects/memory. It has two sections that EVE files automatically:
- Done logs completed work with dates.
- Knowledge holds facts worth keeping, and EVE updates an entry in place rather than adding a duplicate. This section keeps up to 20 entries.
When the Done log grows past its limit, EVE moves the oldest entries into an Archive section. Nothing is deleted. The ledger can also hold a few sections that EVE maintains itself, such as the destination and open decisions.
Periodic Tidying
When you return to a busy space, EVE can tidy the ledger in the background. This happens at most about once a day, and only after at least three new dated Done entries. EVE merges overlapping Knowledge notes into one and moves the originals to Archive. It never blocks your request, and a failed tidy simply does nothing.
This pass maintains what EVE knows. It does not change how EVE reasons or builds.
What EVE Remembers
- Project structure and naming conventions you prefer
- Trade-offs and decisions from prior generations
- Which integrations you authorized and which you rejected
- Errors and how they were resolved, so they do not repeat
Connection to Taskade
EVE Memory ships natively in every Taskade workspace where Taskade EVE is active. The projects/memory folder is visible in your sidebar, and entries link to the Workspace Memory Graph for cross-project visualization. The same surface powers continuity in cloned Taskade Genesis apps.
Why keep memory readable? Recent research (Dream-RSI, arXiv:2609.14858, Google and Google DeepMind, 2026) found that injecting summarized lessons into a prompt underperformed the unguided baseline. EVE Memory takes a different route. It keeps a record you can read and correct, and it does not depend on hidden summaries.
Frequently Asked Questions About EVE Memory
Where does Taskade EVE store its memory?
In your workspace, as regular Taskade Projects in the projects/memory folder. Older workspaces can show projects/memories, and EVE reads both.
Can I edit or delete what EVE remembers?
Yes. Open any memory project and change it like any other project. EVE reads your edits the next time it works in that workspace.
What is the difference between workspace memory and account preferences?
Workspace memory belongs to one workspace and is shared with its collaborators. Account preferences belong to you and apply in every workspace. You edit them under Settings → Account.
Does EVE memory ever delete old entries?
No. When the Done log grows too long, EVE moves older entries to an Archive section. When it merges overlapping Knowledge notes, it archives the originals, so a bad merge stays recoverable.
Does EVE learn to get better on its own?
No. Memory changes what EVE knows about your work: decisions, preferences, and saved facts. It does not change the model or how EVE reasons.
How is this different from a vector database?
A vector database stores embeddings that you cannot read. EVE Memory stores text in projects that you can open, search, share, and correct. See Persistent Memory and Memory Types for the wider landscape.