RESEARCH DIRECTION
Memory, with boundaries.
What should an agent remember, who can access it, and when should it forget?
Explore memory governanceDESKLEMUR LABS
Better AI.
One measured step
at a time.
The research behind DeskLemurOS. Less input to process today, and new questions about how AI could adapt through experience.
01 / MEASURED WITH DPMS V1
Less input.
A lighter start.
Standard uses KV cache. Pro adds DPMS alongside KV cache to reduce input tokens and processing time.
Our re-measured v1 comparison found 85.5% fewer input tokens at the first planner call of each turn. Explore the numbers and the complete-session results.
Explore the measurement02 / DIRECT PLASTICITY
Can AI grow
through experience?
ONGOING RESEARCH
Direct Plasticity explores more adaptive AI: how experience might support useful memory, continuity, and improvement over time.
We are developing this direction in stages. DPMS v1 provides an input-token measurement; later stages need their own evaluation.
03 / THE RESEARCH ROADMAP
Where we are.
What comes next.
- v1Measured
Token efficiency
The re-measured first planner calls used 85.5% fewer input tokens across 14 turns per run.
View the results - v2Development complete
A research milestone
A completed stage exploring memory and continuity for AI.
- v3In development
More adaptive AI
Working toward memory that adapts as tasks and information change.
- v4Planned after v3
Autonomous evolution
A future direction: systems that improve through experience. Work follows v3.
Research status as of October 2026. Development milestones are not release dates or guarantees of public availability. The v1 measurement does not establish the performance of later versions.
THE QUESTIONS BEHIND THE WORK
Built with intention.
PRODUCT PHILOSOPHY
Discovery to engineering.
A capable model is a starting point. Making it an understandable, operable system is the next challenge.
Read the philosophy