Systems research for production agents.
Our research tracks sit upstream of delivery: memory, agent personality, protocol-native tools, eval datasets, and closed-loop knowledge.
Synthetic Personality
Stable beliefs, preferences, memory boundaries, and refusal patterns for long-running agents — measured as behavioral invariants across sessions.
Prior art · Builds on Constitutional AI (Anthropic, 2022), Sleeper Agents (Anthropic, 2024), persona consistency benchmarks, and RLAIF.
Agent Memory
Working, episodic, semantic, and organizational memory with explicit lifecycle controls — write, decay, retrieve, retire.
Prior art · Engages with MemGPT, LangMem, agent-state literature, and the broader long-horizon agent line of work.
Closed-Loop Knowledge
Conversation traces, source graph updates, regression datasets, and flagged corrections in one loop — feedback that updates both knowledge and tests.
Prior art · Adjacent to RLHF data pipelines, automated eval generation (DSPy, MIPROv2), and continuous-evaluation practice.