Hermes Persistent Agent

An investment-research agent that runs on schedule, saves reports, synchronizes versions and can recover after operational failures.

The question

What must change before an AI agent can move from a one-off conversation to an operable and recoverable long-running service?

Approach

  1. Separate scheduled execution from interactive use.
  2. Persist every report and execution state.
  3. Keep deployment and version synchronization observable.
  4. Document recovery steps instead of treating failures as exceptional.

Result

Scheduling, report storage, version synchronization and remote maintenance have run in one environment.

Evidence

Deployment, task execution, synchronization and recovery records.

A useful agent has to survive the conversation

Hermes moves investment research from a one-off conversation into a scheduled cloud service that saves reports, synchronises versions and retains a recovery path when operations fail.

Treat recovery and observability as product capabilities

Only valuable recurring work stays online; deployment follows actual availability; recovery is treated as a product capability; and every completed state must be observable.

From prototype to long-running operation

The project narrowed the persistent-agent scope, changed hosting when capacity was unavailable, connected Discord, memory and scheduling, then added remote maintenance and observable execution records.

  1. Scope the persistent work
  2. Deploy to an available environment
  3. Connect input, memory and schedules
  4. Add remote maintenance
  5. Verify execution and recovery records

The operating record is part of the result

Scheduled research, saved reports and remote maintenance now operate in one environment. Deployment, task execution, synchronisation and recovery records remain available for review.