Blog
Memory, measured.
Case studies and notes from the team building Alexein: what we tested, and what we found.
Looking for the films? They’re on the films page.
- 2026-11-17Right the first time: Claude Code with and without AlexeinTen coding tasks whose answers live in Slack, GitHub, email and meetings, five setups, 150 runs. On the nine knowledge tasks, Claude Code with Alexein made the right change on the first try in 67% of runs, against 15% alone: 4.5× as many. The setting, the setups and every result, with the whole test public.
- 2026-11-17We repeated Mem0’s “97% smaller” Claude Code testMem0 reported that their plugin cut the memory in Claude Code’s context from 13,700 tokens to 445. We ran the same test with Alexein, three times per setup, and counted our own overhead too: on the analysis task, 98% less their way and 89% less ours; every test still passing; and a preference that survives a fresh checkout in 3 of 3 runs.
Put memory to work.
Start with the agents you already use, or add memory to your own app with the API.