Herdr Terminal Agent Runtime
A terminal multiplexer for running, observing, and coordinating coding agents in persistent panes.
Per-dimension scores and reasoning
Evidence shows a terminal workspace manager with CLI and socket API, but no explicit permission model, user confirmation, or data flow. Dependencies are pinned but no security audit. External effects include background server and command execution, but not detailed. Rollback and source attribution not clear. Deductions: lack of explicit least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects control, rollback, and source attribution documentation or implementation evidence.
Project has CI config and tests, but no test results or failure message details. Dependencies pinned but no availability guarantee. Deductions: failure messages not documented, dependency availability not explicit.
README describes multiple use cases and features, but capability boundaries and trigger precision not explicit. Environment fit includes Linux, macOS, Windows, but not detailed. Deductions: trigger precision and some environment fit details insufficient.
README provides install and usage instructions, but lacks FAQ and known limitations. License is Apache-2.0, but version changelog not provided. Maintenance responsibility not clear. Deductions: missing FAQ, known limitations, version changelog, and maintenance responsibility.
Output usability shown via CLI and socket API, but no concrete examples. Marginal value described but not quantified. Cost-benefit not stated. Deductions: lack of concrete examples and cost-benefit analysis.
Claims in README lack specific evidence like test results or performance data. Cross-source verification insufficient. Facts and inferences not clearly separated. Deductions: lack of traceable claims and cross-source verification.
- Unverified publisher identity, proceed with caution.
- Install script via curl pipe, supply chain risk.
- Dependencies pinned but no security audit, verify yourself.
- Documentation lacks explicit permission model and data flow, assess before use.
What does this agent do, and when should you use it?
Herdr is a terminal-based multiplexer and runtime for coding agents, delivered as one Rust binary without Electron. It presents real terminal views for multiple agent panes and surfaces their blocked, working, and done states. Users can split panes, detach while agents continue running, and reattach from another terminal or over SSH; sessions are described as surviving restarts. A pure socket API lets agents create panes, read output, and wait for one another. The project is Apache-2.0 licensed and includes plugins for extending panes and workflows.
Running herdr starts a terminal session in the directory where work lives, with panes in which users run coding agents. It shows live terminal content rather than a wrapped interpretation and accepts both tmux-style prefix keys and mouse actions such as clicking, dragging, and splitting. ctrl+b q detaches without stopping agents in the panes, while running herdr again reattaches. Its socket API allows agents to spawn panes, read output, and wait on other agents.
- A developer coordinating several parallel coding tasks can split a terminal into panes and inspect each agent's live status and terminal view.
- An engineer leaving a long-running task can detach with
ctrl+b qand later reattach from another terminal or over SSH. - A team building coordinated coding-agent workflows can use the socket API so agents spawn panes, read output, and wait for each other.
- A terminal-focused developer who also wants pointer interactions can combine prefix-key controls with mouse-based clicking, dragging, and splitting.
- A team that needs custom pane behavior or workflows can extend Herdr through its plugin system and marketplace.
What are this agent's strengths and limitations?
- Shows real terminal views of agents instead of a wrapped or interpreted output layer.
- Agents continue after detachment and sessions are described as surviving restarts, supporting long-running work.
- Combines tmux-style prefix keys with first-class mouse actions for clicking, dragging, and splitting.
- The pure socket API enables agents to spawn panes, read output, and wait on other agents programmatically.
- Its core interaction boundary is the terminal, so it best fits command-line and pane-oriented workflows rather than a graphical IDE workflow.
- Windows support is explicitly labeled beta.
- The supplied material does not enumerate supported coding agents, so teams need to verify compatibility with their intended agent.
- Socket API details, authentication behavior, and plugin-development specifics are not included in the supplied material.
How do you install or deploy this agent?
A terminal is required. Install with curl -fsSL https://herdr.dev/install.sh | sh. Alternatives are brew install herdr and mise use -g herdr; the documented Windows beta command is powershell -ExecutionPolicy Bypass -c "irm https://herdr.dev/install.ps1 | iex". The supplied installation and local-start instructions do not specify credentials.
How do you use this agent?
Change to the directory where the work lives and run herdr. Start coding agents in panes and split panes as needed. Press ctrl+b q to detach while agents keep running, then run herdr again to reattach. The supplied material also states that reattachment can occur over SSH. The session shows real terminal content and agent state in its panes.
FAQ
Does Herdr write code on its own?
What happens when I detach?
ctrl+b q detaches the session while agents keep running, and the session is described as surviving restarts.