run a Hermes model as a managed employee in a Paperclip company
a lifecycle + orchestration layer — capability still comes from the model and its own harness
lifecycle
activity-based timeout
idle + hard-max + grace, not a dumb wall-clock
smart resume
resume only if the last run ended safely
retry where it left off
before/after session ids in the UI
operations
db-backed usage + cost
real tokens from session db, not regex
per-agent profiles
isolated config, memory, skills, history
agent identity (jwt)
comments post under the agent's name
reach
delivery targets
telegram, discord, slack, email…
memory scoping
session / persistent / ephemeral
session tagging
separate tool runs from user sessions
extensibility
external plugin loading
via adapter-plugins.json
reasoning-effort dial
pass the thinking level through
instruction bundles
managed brief files per role
where it overlaps the goal: this is the “run the model as a managed worker” layer — localharness is the capability inside the worker. · @henkey/hermes-paperclip-adapter 0.4.3