Research for the long horizon.
Senku Labs is an independent research and engineering lab building memory-first AI systems that stay stable as models evolve: durable personalization, safe tool-use, and measurable reliability.
Contact: hello@senkulabs.vn
- Memory
- Evaluation
- Agents
- Systems
Research priorities
- Persistent personalization at scale: instant recall via hot memory packs, with retrieval as an on-demand tool.
- Online memory and long sessions: fast callbacks, streaming-first latency, and robust state handling.
- Reliability as a research object: eval gates, regression suites, and auditable memory (no silent drift).
- Portability: memory that survives model swaps, architecture changes, and enterprise constraints.
What makes Senku Labs different
- We treat memory as infrastructure: versioned, permissioned, and cheap by default (no 10-20s retrieval tax per turn).
- We build for real deployment constraints: latency budgets, migrations, and rollbacks.
- Our symbol is the swallow flying upward: a reminder that the work is long-horizon, open-ended, and always climbing.
Strategy
- Ship small, real systems first. Keep the interface stable. Let models evolve underneath.
- Make memory auditable and reversible (versioned, provenance-backed).
- Prefer fast defaults: no web browsing or vector search on every turn.
Values
- Intellectual curiosity and obsessive experimentation.
- Clear, measurable engineering over vibes.
- Security and governance as product features, not afterthoughts.
Start here
- Blog (updates and notes)
- Research (what we are working on)
- Open Positions (if we are hiring)