Quick answer
Letta is a stateful agent platform with editable persistent memory, git-backed MemFS, human interfaces, and an official hosted MCP server. That server lets another assistant create and message Letta agents. Hjarni is a Markdown knowledge base whose official hosted MCP server lets connected assistants search, read, and update notes directly. Choose the runtime when you want to run agents, and the knowledge base when you want shared documents for the assistants you already use.
Details behind the comparison
- Primary shape
- Letta runs agents; Hjarni supplies knowledge to connected assistants.
- Memory control
- Letta memory is not owned exclusively by an agent. Developers can edit blocks, and current agents use editable MemFS files.
- MCP server
- Letta exposes a hosted server for creating and messaging agents and reading replies. This is agent delegation rather than direct note CRUD.
- External tools
- Letta can both serve an MCP endpoint and call external MCP servers.
- Human interface
- Letta provides interfaces for people to interact with agents and their memory.
- Versioning
- Letta commits memory edits to a git repository. Compare the actual review and restore workflow, rather than assuming version history is absent.
- Shared context
- Both support shared knowledge, through different product models and plan requirements.
- Self-hosting
- Letta documents self-hosting; deployment responsibilities depend on the version and setup.
- Instructions
- Hjarni folder instructions guide connected assistants; they do not guarantee tool calls or recall.
- Pricing
- Letta has a hosted Free plan, Pro at $20/month, Teams Pro at $20/seat/month, and an API plan with usage charges.
An agent platform and a knowledge base
Letta provides stateful agents through a web app, CLI, SDK, and related tools. Hjarni provides Markdown notes, folders, tags, team sharing, and an editor. Both can hold material a person edits, and both can receive AI changes.
Letta has an official hosted MCP server
The hosted Letta endpoint accepts an API-key bearer token. Its tools list and create agents, send messages, and fetch run status and replies. It requires a client that supports custom bearer headers; OAuth-only clients are not supported by that endpoint. Letta agents can also use external MCP tools.
Memory can be inspected and edited
Letta's current MemFS stores memory in a git-backed repository, with commits for edits and synchronization for cloud-backed agents. Its earlier block-based API also permits direct developer edits. Hjarni exposes note versions and one-click restore in the app. Neither design removes the need to review AI-written content or verify that an assistant used the right context.
Pricing and fit
Letta offers hosted Free, individual Pro, Teams Pro, and developer plans. The developer API has usage charges for active agents, tool execution, and model use. Self-hosting is a separate deployment choice. Hjarni uses flat knowledge-base subscriptions with note quotas; your assistant subscription or provider usage remains separate. Choose Letta for persistent agents and Hjarni for documents shared across connected tools.
Letta runs persistent agents. Hjarni keeps the knowledge your assistants use.