Migration guide
Migrating from Letta to Cognee
Letta’s platform changes left teams that built on its memory APIs looking for a maintained home for agent memory. This guide outlines how to move that memory to Cognee’s open-source platform.
Based on public information as of June 2026. See each vendor's site for current details.
In March 2026, Letta (formerly MemGPT) deprecated Letta Cloud and the Agent Development Environment and pivoted to agent products. Teams that built on its memory APIs need a maintained home for agent memory — and ideally one where a future pivot cannot strand them again.
Why teams migrate to an open-source core
Cognee’s core is open source. Memory lives in your infrastructure and the code stays available, so a vendor’s product pivot cannot take your memory layer with it.
Your data is stored in systems you control and remains exportable. Run fully local, self-host, or use Cognee Cloud — the choice stays yours over time.
Cognee has 17k+ GitHub stars and 5M+ SDK runs per month, and runs in production at Bayer, the University of Wyoming, and Knowunity.
Migration outline
- 1
Export your memory from Letta
Pull your memory blocks and archival data out of Letta while you still have access to them.
- 2
Install Cognee and configure storage
pip install cognee, then point it at the graph and vector storage you want to use — fully local works.
- 3
Re-ingest your data
Load the exported memory with cognee.add(), then run cognee.cognify() to build the knowledge graph and vector index.
- 4
Point your agents at Cognee
Query memory with cognee.search() from the Python SDK, or connect Claude Code, Cursor, and other MCP clients through the MCP server.
The exact mapping depends on how your agents used Letta’s memory blocks, so treat this as a starting outline rather than a script. The full setup and API reference is at docs.cognee.ai.