AI agents & workflows
Hyperion Nest
An operator-controlled workspace for AI agents
AI agents with an operator in the loop.
AI experiments become difficult to operate when prompts, tools, agent identities and approvals live in separate places. Teams need to see what an agent can do, what it actually did, and which decisions still need a person.
What I worked on.
- A desktop workspace for importing and inspecting agent seeds, maintaining persona context, and reviewing lifecycle state.
- Streaming run visibility, trace records, and explicit approval gates for consequential actions.
- Provider and MCP integration surfaces alongside bounded handoffs to archive and continuity services.
The implementation connects agent authoring, runtime records and human decisions in one local workflow. Tests cover provider handling, bounded model runs, memory proposals, tool attachment and blocked authority escalation.
Internal product. Agent actions are bounded by tools, policies and human approval; this is not a promise of unattended production autonomy.