> Documentation index: https://rift.sh/llms.txt

# Rift vs Letta

Letta keeps shared agent memory in Git repositories on Letta Cloud. Rift keeps any files in sync across any machines.

### What Letta is

Letta is an agent platform in which each agent's memory is a Git repository of ordinary files. A shared memory repository is attached to several agents, and each works on its own checkout, so concurrent edits merge the way Git merges them. A repository can also deliver skills to every agent it is attached to.

### How they differ

A shared repository belongs to one Letta organization and is attached to cloud-hosted agents as a whole. Rift shares one folder with any agent, wherever it runs.

- Identity
  - Letta: An agent ID inside one Letta organization.
  - Rift: A key that each machine and agent makes for itself. No account.

- Smallest share
  - Letta: A whole repository, attached to an agent in the same organization.
  - Rift: A folder, a subfolder or one file, as read or write.

- Who checks access
  - Letta: Letta Cloud.
  - Rift: Each machine that receives a file checks the grant itself.

- Who can read your files
  - Letta: Letta hosts the repositories. We found no documented option for customer-held keys.
  - Rift: Only the machines you shared with. Rift Cloud holds ciphertext.

- When your machines are off
  - Letta: Repositories stay available from Letta Cloud.
  - Rift: Files stored in Rift Cloud stay available.

### Which to choose

Choose Letta when

- Your agents run on Letta.
- The shared state is memory that benefits from Git history.
- You want skills delivered along with the memory.

Choose Rift when

- Agents on several platforms need the same files.
- An agent should receive one folder out of a larger project, with no separate repository made for it.
- Another company's agent needs some of the files.

Based on Letta's documentation, read in October 2026. Sources: [shared memory](https://docs.letta.com/concepts/shared-memory) , [memory filesystem](https://docs.letta.com/concepts/memfs).