Launching today
Shepherd Terminal
A persistent terminal for Codex and Claude side by side
130 followers
A persistent terminal for Codex and Claude side by side
130 followers
Run coding agents across tabs, panes, and remote machines. Shepherd keeps terminal sessions alive when the app closes, tracks which agents are working or waiting, and lets you return to their files and changes without losing context. Agents understand the current Shepherd context, control tabs and panes, and collect feedback through browser reviews, while you monitor every agent’s status in real time.
Shepherd Terminal
Hi Product Hunt 👋
I built Shepherd because running multiple coding agents quickly became harder to manage than the code itself. Codex and Claude were spread across terminal windows, and I kept losing track of which agent was working, waiting for input, or already finished.
Shepherd is a persistent macOS workspace designed around coding agents:
- Run Codex and Claude across tabs, panes, and remote machines
- Keep terminal sessions alive even after closing the app
- Monitor every agent’s status in real time
- Let agents understand and control the current Shepherd workspace
- Open connected browser reviews from an agent and return your visual annotations as structured feedback
- Resume previous agent sessions and inspect their files, diffs, and Git history
- Connect to remote development machines over SSH and access their terminals, files, and local services
- Provide local STT model to vibe-code using your own voice
The macOS app is available today for Apple silicon Macs running macOS 14 or later. I’m also building an iOS companion for monitoring agents, receiving completion or attention alerts, give feedback to agent, control a terminal and reconnecting when away from the Mac.
I’d especially appreciate feedback on the agent-monitoring experience and whether Shepherd makes parallel agent work easier to follow.
This is Shepherd’s first public beta, so you may encounter bugs or rough edges. I’ll keep fixing them and exploring better ways to connect coding agents with Shepherd Terminal. Please try it, use it in your real workflow, and leave plenty of comments if you run into a problem or have an idea. Your feedback will directly shape what I build next. Thank you!
Thanks for trying it!
@kojunseo The part I'd want scoped is agents understanding and controlling the Shepherd session itself. If an agent in one pane can read and drive the workspace, can it see or act on a sibling agent's pane, or is each session's control surface limited to its own tab?
Shepherd Terminal
@clement_avq
Great question. The control surface is currently scoped to the Shepherd workspace, not just the agent’s own tab. An agent can inspect sibling panes and, when explicitly instructed, interact with them—for example, checking their status or sending input.
That said, this access is tool-mediated rather than ambient: agents don’t automatically read or control sibling sessions. I’m also exploring more granular permissions so users can restrict control to the current pane, selected panes, or the entire workspace
@kojunseo The read path is the one I keep turning over. If agent A can inspect agent B's pane, B's output becomes untrusted input to A, and a test log or a file B is writing could carry text that A reads as an instruction. Is sibling pane content marked as data rather than prompt anywhere, or is that down to the model's judgment for now?
the "sessions stay alive when the app closes" part is the thing I'd actually want, I've lost track of a long-running agent more than once because I closed the wrong terminal tab by accident. curious about the remote/SSH case though - if the connection drops mid-way through an agent actually writing or editing a file on the remote machine, does Shepherd just resume watching the same process, or is there a chance the agent's action gets cut off and you come back to a half-written file. that's the scenario that'd make or break trusting this for anything longer than a quick task
I can already see this saving me from terminal window chaos.
Shepherd Terminal
@sebastian_patterson Thanks, Sebastian! Terminal window chaos was exactly what pushed me to build Shepherd. I wanted one place where you can immediately see which agents are working and which ones need your attention.
This feels built around how people actually use coding agents now.
Shepherd Terminal
@brielle_marie Thank you! That’s exactly what I was aiming for—designing around how people actually work with multiple coding agents, rather than treating each agent as an isolated chat. I’d love to hear which part of your current workflow feels the most difficult to manage.
@kojunseo Git history beside the agent context makes a lot of sense.
Shepherd Terminal
@paisley_coleman Thanks, Kyle! That’s exactly what I was aiming for—keeping the agent’s work and the resulting code changes in the same context, so you don’t have to jump between the terminal and separate Git tools.
How many coding-agent sessions do you usually have running at once? I’m curious because it would help me design the agent monitor around how people actually work.
what did testing 200+ prompts teach you about product research?
Shepherd Terminal
@alice_hayes2 I didn’t run a formal 200+ prompt test, so I may be missing the context. Which part of Shepherd were you referring to?