MemoryRouter vs Letta
Letta is an agent platform where the agent owns its memory blocks. MemoryRouter is a memory layer that follows you across tools. One is a runtime; the other is a vault.
Competitor details and pricing checked 2026-09-25. Sources at the bottom of this page.
What Letta is
Letta (formerly MemGPT) builds stateful agents whose memory is part of the agent state, with the agent editing its own memory blocks. Published plans: Free at $0 with limited agents and bring-your-own-keys, Pro at $20 per month with up to 20 stateful agents, usage-based developer plans for API workloads with unlimited agents, and team or enterprise plans with shared resources.
The model is agent-centric: memory belongs to an agent you run, and it is edited as part of the agent loop.
What MemoryRouter is
MemoryRouter is memory-centric: a vault that multiple tools read and write. Claude Code and Codex capture turns automatically through hooks, editors connect over MCP, and the vault is the same one everywhere.
14-day free trial, then $20 per month with 200M memory tokens, unlimited projects, no inference markup.
Side by side
| MemoryRouter | Letta | |
|---|---|---|
| Core model | Stateful agents with self-edited memory blocks | Vaults read and written by many tools |
| What owns memory | The agent | The vault, independent of any agent |
| Entry price | Free tier with limits, Pro at $20 per month, usage-based developer plans | $20 per month after a 14-day trial |
| Coding-tool capture | Through the agent you build or run | Hook packages for Claude Code and Codex, MCP elsewhere |
| Cross-vendor continuity | Wherever the agent is deployed | One vault across vendors and surfaces |
| Open source | Yes, open-source agent framework | Published clients, managed service |
Where Letta is the better choice
- You want to build or run an agent that maintains its own memory as part of its state.
- You want an open-source runtime you can inspect and extend, with a cloud option.
- You are comfortable with agents as the unit of deployment rather than memory as a shared service.
Where MemoryRouter fits better
- You want your existing tools to remember, without running an agent platform.
- You want memory that outlives any agent runtime and is readable by tools the runtime never sees.
- You want capture that happens automatically on completed turns instead of being a decision the agent makes.
Which one should you use?
Letta is compelling when agents are the product, especially if you want the memory architecture to be part of the agent design.
MemoryRouter fits when the product is your existing workflow and the problem is that it forgets. Vault first, tools second, no runtime to operate.
One vault for Claude Code, Codex, Cursor, Claude, and ChatGPT. 14 days free.
Related
FAQ
Sources
Sources checked 2026-09-25. Tool behavior changes; check the vendor docs before relying on a detail.