Skip to main content
For AI agents: a documentation index is available at https://docs.parallel.ai/llms.txt. The full text of all docs is at https://docs.parallel.ai/llms-full.txt. You may also fetch any page as Markdown by appending .md to its URL or sending Accept: text/markdown.
Memory lets your agents build on research you’ve already done with Parallel. When you use Parallel personally — through the CLI, or third-party agents like Claude Code — your agents can search past Task, Monitor, and FindAll results and reference them in new work. This personal memory never changes the results of Task, FindAll, Monitor, or other API requests.
We’re gradually rolling out memory over the coming weeks. As part of this rollout, memory will be enabled by default for some newly created organizations.

​
How memory works

Once memory is enabled, the inputs and outputs of supported Parallel runs are saved to your personal memory. When your agent starts new research, it can choose to search your memory for relevant past work. Memory is never used automatically — your agent must explicitly retrieve it. Only you and your personal agents can access your personal memory, and it follows you across the agents that you use.

​
Memory example

​
Turn personal memory on

Members can turn on personal memory under Settings → Account once memory is enabled for their organization. During opt-in, you’ll be shown the account keys that will contribute to your personal memory. When memory is turned off, new memories will not be created. Existing memories are retained and become available again if memory is turned back on. You can also permanently delete your personal memory. This cannot be undone.

​
Organization controls

Organization admins manage memory under Settings → Organization. The organization setting controls whether members can use personal memory. Each member must still opt in individually. Turning memory off stops new memories from being created. Existing personal and application memories are retained until they are permanently deleted. Admins can permanently delete all personal and application memory. This cannot be undone.

​
Taking advantage of memory

Use the memory API to take advantage of memory. Access the API using the Parallel CLI or the Parallel skill. Update to the latest version of the Parallel CLI:
Update to the latest version of the skill:
Ask your agents to “check what I’ve already researched with Parallel before starting.”

​
Building applications with memory

If you’re a developer building on Parallel, you can also create application-scoped memory. Application memory is opt-in per request. Runs are saved only when your application provides a memory_scope_key. Each memory_scope_key maps to its own isolated memory bank. You can set memory_scope_key when creating a Task run, a FindAll run, or a Monitor.

​
API reference