I build runtime accountability infrastructure for long-lived AI-agent systems: a self-hosted checkpoint layer that operates after deployment, while agents are running. At meaningful checkpoints it binds claims to a process identity, updates a longitudinal state estimate, returns an auditable proceed or pause action, and preserves the evidence behind it.

The maintainer deployment has run continuously on a single-operator development fleet since November 2025. That establishes sustained operation under real load, not incident prevention, predictive utility, or cross-operator generality. External pilots are the next step.

Papers, systems, datasets, and decks, all in one place. That page is canonical; this profile is just the front door.

Start here:unitares → docker compose up -d --wait && make demo drives a synthetic agent through six check-ins and prints the policy response at each step.

Build on it:unitares-sdk on PyPI is the agent-side contract (check-in loop, EISV state, identity anchors). pip install unitares-sdk.

• What this is: self-hosted runtime accountability and state estimation for long-lived agent processes, the checkpoint layer between evals/guardrails and incident response.
• What this is not: not an output filter, not a sandbox, not a correctness or ethics oracle, and not evidence yet of preventive benefit or cross-operator generality.
• Current ask: external pilots and design partners who already run autonomous agents long enough for drift, calibration, and recovery to matter.