IGRIS

Agent Orchestration

You send a request from any chat app. Hermes, your assistant, asks the decision layer (Laya or Jev) to pick the right agent, model and risk level. A builder agent writes the code and a reviewer agent checks it, each in its own Herdr pane, and either one can ask the decision layer mid-task. The shell runs the tests, and Hermes reports back to you. Pick your own tools in Your setup, and hover or tap any part to see what it does.

~/app · herdr
builds, reviews, decides
hermesherdr
HERMES
assistant · always on
new request
reads your message,
splits it into tasks
task:
» waiting
ask
model any cheap llm
tasks done 0
cost today
$0.00
dispatch
never writes code
itself. it hands
work to herdr panes
trusts the shell,
not an agent's
word for it
verify in shell
done = tests pass
+ review approved
merge + deploy
stay with you
report to you
YOU ·
» …
· decision layer · decisions 1,204
which agent——
model tier——
risk——
sharp → hermes acts · split → stronger model or you
dispatch into herdr panes · hermes
herdr
workspaces
~/app
~/pos
agents
shell
builds · —
0%
waiting
idle
reviews diff
0%
waiting
idle
shell
tests · git
0%
waiting
idle
verify + report
HERMES → YOU · report
nothing to approve