Human-in-the-loop AI 4

Personal AI assistant that coordinates work, agents, health data, and daily workflows. Waldo is a personal AI action layer that coordinates agents, tools, apps, health signals, and personal context to manage work and daily life. Its first surface, Kennel for Mac, helps users define goals, coordinate multiple AI sessions, delegate tasks to agents such as Codex, Claude, and OpenCode, and return only decisions that require human judgment. Waldo also plans daily workflows, analyzes health and recovery data, drafts communications, tracks changing patterns, and verifies what has been completed.

AI workforce platform for managing teams of specialized agents Aldena is an AI workforce platform that lets businesses create dedicated rooms staffed by specialized AI agents. Users can define agent roles, build reporting hierarchies, set goals, and allow managers to delegate work while maintaining shared room context and private agent memory. Agents operate through connected business tools such as GitHub, Bitbucket, Jira, Linear, Notion, Google Drive, Gmail, Slack, and Vercel. They can produce pull requests, deploy applications, write documentation, update backlogs, and complete other delivery tasks, while human users review and approve actions before anything is merged, deployed, or shipped.

AI grading assistant for instructors to streamline feedback and grading processes with human oversight. TimelyGrader is an AI grading assistant that empowers instructors to streamline feedback and grading processes. It integrates AI with human controls at every step, ensuring accuracy and consistency. TimelyGrader supports various assignment types, including essays, papers, presentation decks, and spreadsheets. It provides scalable AI feedback, reduces grading time, and helps instructors give more feedback to students without extra work.

Machine-native AI for reliable, calibrated software decisions TypeSafe AI is an AI research lab building machine-native intelligence infrastructure for software automation. Its first public System One Model, Jev, produces typed decisions with calibrated probabilities and confidence estimates rather than conversational text. Jev is designed for reliable, fast, and cost-effective automation, allowing software to act autonomously when confidence is high and escalate uncertain decisions for human review.