Herdr vs Hermes Agent: Terminals or Persistent Agent Work?

6 min read Comparisons

Herdr supervises terminal agents. Hermes Agent adds persistent memory and profile-based Kanban. Compare recurring work, approvals and checks.

Herdr and Hermes Agent both let you run agents on your own machines, but do you need to supervise terminal programs or preserve agent work across sessions? For a developer repeating investigations or maintenance, the deciding difference is whether a tab and its terminal history are enough to carry the next occurrence.

Herdr starts existing terminal agents unchanged and lets a parent collect reports from helpers. Hermes Agent publishes persistent, editable memory alongside cards assigned to agent profiles. Both coordinate agents, but their published records make different promises about what survives into later work.

Herdr keeps the terminal as the working surface

Herdr's guide organizes the desktop around spaces, tabs and terminals, suggesting a space for a repository and a tab for a task. An agent can start in a worktree for an issue. Helpers can perform a test run or review and return their reports to the parent.

That suits a developer who already trusts particular terminal programs and wants to follow several of them together. Herdr runs the agents using their own model accounts, and connects to your other machines over SSH. Its core documentation describes the runtime and API that clients use to start or message agents.

The mobile interface sorts agents by whether they need you, are done, are working or are idle. You can answer prompts inside their real terminals. This is useful remote supervision, with a specific boundary: answering a terminal request does not establish a permission policy enforced by Herdr.

Hermes Agent retains notes and profile-owned cards

Hermes Agent's memory guide describes persistent notes and preferences for later sessions. You can inspect, change or remove learned memory through journey commands or the desktop panel. A correction has a documented home that you can review before it informs later work.

Its Kanban documentation assigns cards to named profiles and preserves comments, dependencies and attempt history. An orchestrator can own the root task, delegate specialist cards and judge whether the project is complete. People manage and comment on cards, but their documented assignee is an agent profile.

That distinction is important for a small team. Agent coordination is published, while shared human-and-agent assignment is a separate requirement. The board also uses a trusted local-user model and is single-host by design, so evaluate that boundary before making it your team's common work record.

Memory and recurrence must reach the next run

Herdr core retains terminal history and opaque API documents, according to its security documentation and schema documentation. Its public pages do not describe editable agent memories that improve later work. The terminal agent you install may have its own behavior, but that should not be attributed to Herdr's runtime.

Hermes Agent explicitly carries persistent notes into later sessions. For a recurring investigation, inspect whether the note records a useful correction, such as a repository convention, and whether you can remove it when that convention changes. Retained output and a maintained lesson serve different purposes.

Herdr's mod instructions allow repeated code through cx.timer.every. Its public pages do not describe outside-event or webhook setup for starting recurring work. Hermes Agent's cron guide and webhook guide document recurring jobs and external events that can start them immediately.

Both require attention to the runtime. Herdr says a machine reboot stops agents, and restored panes return as shells without resuming the work. For self-hosted Hermes Agent, agent-backed scheduled jobs require a running gateway, and a scheduler tick skips them when it is down. Write a recovery procedure for whichever process must stay available.

A terminal answer and an approval policy differ

Herdr core explicitly says it is not a sandbox or permission broker. A caller connected to its API can run commands and change files with the account's access. Its mobile Inbox helps you find and answer prompts, but its public pages do not describe configurable action approvals enforced by Herdr itself.

Hermes Agent's security guide documents shell approval modes, permanent allowlists and deny rules. Dangerous-command requests can be approved or denied in the terminal or messaging chat. Its approvals suggest command also turns executed, approved commands into ranked allowlist proposals, excluding destructive command classes and leaving application to you.

For an unattended maintenance job, inspect the command boundary before the first run. Decide what should always need a person, then test a request at that boundary. A notification surface tells you which agent wants attention, while an approval policy determines whether the requested command may run.

Completion evidence and hosting need explicit contracts

Herdr can run test commands and delegate a separate review. Those are checks you arrange through the terminal programs, not a Herdr acceptance-review gate. Its saved reviewer templates can ask for bugs with paths, lines and reasons, but its public pages do not describe reviewing against configured repository rules.

Hermes Agent's goal documentation provides completion contracts and shell quality gates that must pass before a goal completes. Its Kanban review skill can match acceptance criteria to evidence and return work for correction. However, its documented kernel does not execute workflow-template routing fields, which are reserved for a future version.

Herdr's app and runtime are free and open source, and its official mobile app is free. You provide the terminal agents and machines. Hermes Agent's MIT core is also free, with optional hosting available through Hermes Cloud. Portal lists Plus $20/month, Super $100/month and Ultra $200/month with top-ups, while the Cloud FAQ separately meters compute and storage without explaining how that relates to included hosting. Confirm the configuration's billing terms before choosing it.

Herdr and Hermes Agent, point by point

The table puts both products against the same questions, with the source and check date attached to each answer. It separates a terminal runtime's capabilities from the persistent agent features published by Hermes Agent.

Both columns answer the same questions from each company's public pages. Yes means the company publishes the capability the question asks about, Partly means something narrower, No means it says it does not offer it, and Not published means its public pages do not describe it. Every answer links to its source and shows when we last checked it. The oldest check on this page was November 13, 2026.

Herdr and Hermes Agent compared, with the source and check date of each answer
What you compare Herdr Hermes Agent
Goals Herdr Partly

A community Claude goal mod writes a /goal for your current Claude Code session. Its listing does not describe tracking a measurable target and reviewing progress against it, and Herdr does not review gallery mods.

herdr.dev/…/herdr-mods--claude-goal
Hermes Agent Partly

Persistent goals keep an agent working toward an objective, with a judge and optional completion tests. Its docs describe no linked metric tracking against an ongoing business goal.

Leads Herdr Partly

A parent agent starts helpers, gives them work and receives their reports. Its docs do not describe a project, goal or team lead role that reviews and coordinates the work.

herdr.dev/core
Hermes Agent Yes

You can name an orchestrator profile to own a decomposed project's root task, assign child tasks to specialists, judge completion and add more work when needed.

Tasks Herdr Partly

Herdr suggests a tab per task and can start an agent in a worktree for an issue. Its docs do not describe shared work records that both people and agents can own with their own discussion or history.

herdr.dev herdr.dev/guide
Hermes Agent Partly

Kanban cards keep an assigned agent profile, comments, dependencies and attempt history. People can comment and manage cards, but the documented assignee is a profile name rather than a person or an agent.

Recurring work Herdr Partly

Custom mods can repeat code on an interval with cx.timer.every. Its docs do not describe an outside-event or webhook setup that starts recurring work.

github.com/…/AGENTS.md
Hermes Agent Yes

Cron jobs run once or recur on a schedule, and webhook events can fire the same jobs immediately. Each job can carry its own instructions, skills and delivery settings.

Workflows Herdr Partly

Templates save an agent command and role, and agents can start helpers, wait and read their results. Its docs do not describe a defined multi-step process that runs the same way each time.

Hermes Agent Partly

Kanban dependency chains coordinate roles such as scout, editor and writer. They provide ordered handoffs, but the kernel does not execute workflow templates, whose routing fields are reserved for a future version.

Autonomy Herdr Partly

The mobile Inbox lists agents needing you and lets you answer their terminal prompts. Core explicitly is not a permission broker for agents, and its docs do not describe configurable action approvals in Herdr itself.

Hermes Agent Yes

You configure shell approval modes, permanent allowlists and deny rules. Dangerous-command requests appear in the terminal or messaging chat, where you can approve or deny them.

Learned autonomy Herdr Not published

Its public app and runtime docs do not describe proposing changes to an agent's permissions from your approval history.

Hermes Agent Yes

The approvals suggest command mines executed, approved commands into ranked allowlist proposals. You choose which entries to apply, and destructive command classes are excluded.

Memory Herdr Not published

Core retains terminal history and opaque API documents. Its docs do not describe agent memories that improve later work and can be inspected or corrected.

Hermes Agent Yes

Persistent notes and user preferences carry into later sessions. You can inspect, edit or delete learned memory through the journey commands or the desktop panel.

Planning Herdr Partly

The community Plans layout displays repository plan files, tickets and the agents working on them. Its listing does not establish planning in conversation before execution, and Herdr does not review gallery layouts.

Hermes Agent Yes

From chat, Kanban can expand a rough idea into a specification or dependency-linked tasks. You can disable automatic child promotion so the generated tasks wait for manual review before execution.

Intake Herdr Yes

Custom clients can use core's API to start or message agent terminals. The caller selects the terminal or agent template that receives the work, rather than sending it to a shared business-request queue.

Hermes Agent Yes

Signed webhooks turn outside events into agent prompts. A route can target a named profile when profile multiplexing is enabled, and filter which events start work.

Connectors Herdr Partly

Mods can fetch URLs, and the guide shows a custom ntfy notification. Its docs do not describe MCP connectors or outside-event triggers alongside custom HTTP access.

Hermes Agent Yes

A curated Model Context Protocol catalog and custom local or remote servers connect external tools. HTTP webhook routes accept signed service events and can post responses back through integrations such as GitHub.

Models Herdr Partly

Agent templates put model choices in the command that launches your installed agent. Its docs do not describe Herdr selecting a model automatically per task or step.

github.com/…/templates.md
Hermes Agent Partly

You choose the main model and can override auxiliary tasks. Provider routing chooses a backend for a selected model, rather than selecting a model from each task's needs.

Cloud or your own computer Herdr Partly

Agents run on your computer or machines you connect over SSH, including servers, virtual machines and cloud sandboxes. Its current app and runtime docs do not describe a Herdr-hosted execution option.

herdr.dev herdr.dev/core
Hermes Agent Yes

Run the free agent locally through its desktop or terminal interfaces, or choose the optional Hermes Cloud service for a hosted instance with a persistent workspace.

Your own subscription Herdr Yes

Herdr runs your existing terminal agents unchanged, including Claude Code. You use those programs' own model access and accounts rather than a Herdr model subscription.

Hermes Agent Yes

Hermes accepts your provider API keys and a ChatGPT account login. Its Claude account path requires Max plus purchased extra-usage credits, not the included Max allowance.

Pricing Herdr Partly

The app and herdr-core are free and open source, and the official mobile app is free too. Its pages do not publish usage-credit pricing or an included model allowance.

herdr.dev herdr.dev/mobile
Hermes Agent Yes

The MIT agent is free. Portal lists Plus $20/month, Super $100/month and Ultra $200/month, with credit top-ups. Cloud's FAQ separately meters compute and storage, excluding inference and tools, without explaining how that relates to included plan hosting.

Correctness Herdr Yes

Core runs test commands such as cargo test, and an agent can delegate a test run and read its result. These are checks you arrange through terminal programs rather than a Herdr acceptance-review gate.

herdr.dev/core herdr.dev
Hermes Agent Yes

Goal completion contracts define verification criteria, and shell quality gates must pass before a goal completes. Kanban review also maps acceptance criteria to evidence and can return work for correction.

For software work

Herdr and Hermes Agent compared, with the source and check date of each answer
What you compare Herdr Hermes Agent
Code review rules Herdr Partly

A saved reviewer template instructs an agent to read code and report bugs with file paths, lines and reasons. Its docs do not describe reviewing code against configured repository rules.

github.com/…/templates.md
Hermes Agent Yes

The pull-request review guide uses reusable review guidelines and remembered team conventions, such as required type annotations and database access rules, to check later diffs.

Adversarial review Herdr Partly

An agent can hand a diff to a separate reviewer, and the guide shows a reviewer reporting back. Its docs do not establish a distinct independent review for risky changes.

herdr.dev herdr.dev/guide
Hermes Agent Yes

A bundled pre-commit skill delegates security and logic review to a separate, diff-only reviewer in interactive sessions. Kanban's review skill independently verifies handoffs and keeps the reviewer from editing the implementation.

Risk rating Herdr Not published

Its public app and runtime docs do not describe rating the risk of a code change before merge.

Hermes Agent Not published

Its docs describe no change-level risk rating before merge. Optional residual-risk notes on Kanban handoffs and severity labels for individual review findings do not establish such a rating.

Pull requests Herdr Partly

The community PR Reviews mod checks out a selected pull request in a worktree and starts your agent with a configured prompt. Its listing does not establish agents addressing review feedback, and Herdr does not review gallery mods.

herdr.dev/…/herdr-pr-reviews
Hermes Agent Partly

Agents can open pull requests through the GitHub CLI, and the bundled GitHub skill requires reading PR comments. The cited docs do not establish a review-feedback revision cycle.

A fact we cannot find or cannot check is left out, and prices can change after the date shown. Task Machine publishes this comparison and is an interested party. Herdr and Hermes Agent are named only to identify them, and there is no relationship between the companies. Has something changed? Tell us.

Who each one suits

Pick Herdr if your priority is supervising existing coding-agent terminals, customizing their desktop layout and answering their prompts remotely. Core uses your account's access, and a reboot does not resume agents, so permissions and recovery remain explicit responsibilities.

Pick Hermes Agent if persistent lessons, agent-profile cards, recurring jobs and completion gates match your maintenance process. Its board is single-host, and reserved workflow-template fields do not define an executable workflow in the current documented kernel.

Where Task Machine fits

Task Machine is relevant when the recurring job needs a person or agent to own the same kind of work record. Its Tasks keep the plan, discussion, decisions and runs together, and its Workflows include defined branches, human questions, approvals and verifier checks. Editable memory is scoped across agents and work contexts, with relevant notes carried into later work.

Task Machine is the wrong pick when a free terminal environment or a self-hosted profile-based agent already meets the requirement. Compare Task Machine vs Herdr and Task Machine vs Hermes Agent before adding another work system. Next, document one recurring job's saved lesson, required approval and completion evidence, then verify where each record lives.