Hermes Agent Goals (NEW Persistent Update FREE)

The Hermes agent goals update is the biggest free upgrade to any open-source AI agent in 2026 — and this guide breaks down exactly what it does, how to use it, and the use cases worth building.

This post covers:

What Hermes goals does.
How to install + use.
The four commands (set, status, pause, resume, clear).
Best use cases.

🔥 Want my Hermes goals 30-day roadmap + use cases? AI Profit Boardroom has the full goals roadmap + weekly live coaching. → Get the roadmap

What Hermes Agent Goals Does

One command.

Hermes chases the goal.

Step by step.

Until completion.

Inspired by Ralph Loops (popular with Claude Code).

A judge model checks after each turn whether the goal is done.

If not, the agent continues.

Loop continues until done OR turn budget exhausted OR you pause/clear.

How To Install

In terminal:

hermes update

Pulls latest features including persistent goals.

The Four Commands

1 — `/goal <text>`

Sets a persistent goal.

Example: /goal build out an SEO website with 10 articles.

2 — `/goal status`

Shows current state of the active goal.

3 — `/goal pause` / `/goal resume`

Pause + resume.

You can close terminal, come back tomorrow, resume.

4 — `/goal clear`

Ends the goal.

Watch The Walkthrough

For Hermes basics:

How The Loop Works

Three layers.

Layer 1 — Goal accepted

Hermes confirms the goal.

Initialises planning.

Layer 2 — Turn loop

Each turn = a normal Hermes action.

Code, file edit, web search, etc.

Layer 3 — Judge check

After each turn, judge model evaluates: "Is the goal done?"

If yes → "Goal achieved."

If no → continue.

Turn budget

Default 20 turns.

Configurable in config.

Prevents runaway loops.

Configurable Settings

In config:

goal_max_turns — max turns before stopping.
goal_judge — provider + model used to judge completion.

Bigger budget for bigger goals.

Stronger judge model for nuanced goals.

Best Use Cases For Hermes Goals

Six high-value use cases.

1 — Build full websites

/goal build out a 10-page SEO website on [topic] and deploy to Netlify.

2 — Multi-step research

/goal research [topic] across 10 sources + write a comprehensive summary.

3 — Code refactors

/goal refactor this codebase to use TypeScript + add tests.

4 — Content series

/goal create a 7-day email series on [topic].

5 — Data analysis

/goal analyse this dataset + produce a structured report.

6 — Bug squashing

/goal find and fix every bug in this repo.

For each, one command kicks off autonomous multi-step work.

What Makes A Good Goal Prompt

Three rules.

1 — Specific

"Build a website" → vague.

"Build a 10-page SEO website on Hermes agent topic, dark theme, deploy to Netlify" → specific.

2 — Measurable

The judge needs criteria.

"With at least 10 published articles + sitemap.xml live."

3 — Bounded

Reasonable scope for the turn budget.

If 100 turns needed, set the budget.

Resume Across Sessions

Powerful feature.

Set goal Monday.
Close laptop Tuesday.
Wednesday: /goal resume.

Continues exactly where left.

For multi-day projects, this is the unlock.

Common Goals Mistakes

Three.

1 — Vague goals

Judge struggles to determine completion.

Be specific.

2 — No turn budget for big goals

Default 20 turns won't finish a 50-step task.

Adjust config.

3 — Ignoring goal status

Check status periodically.

If looping unnecessarily, clear and re-prompt.

Pairing With Other Hermes Features

Three combinations.

Goals + skills

Goal triggers skills.

Skills handle specific steps.

Goals + memory

Memory persistence (see OpenClaw Memory Persistence for related concepts) keeps long-running goals coherent.

Goals + multi-agent

Manager agent sets sub-goals for worker agents.

See Hermes Agent Swarm.

Cost Of Running Goals

Local Hermes via Ollama: free.

Cloud LLM-backed: per-token cost.

For typical goal: £0.50-£5 per long-running goal.

For multi-day goals: £5-£20.

Cheap relative to value.

Real Goals I've Run

Five.

Goal 1 — built this site

Goal: 25-article SEO site.

20 turns to first version.

40 turns to polished + deployed.

Goal 2 — research week's AI news

Daily.

5 turns each.

Goal 3 — code refactor

Bigger codebase.

50+ turns over 2 days.

Goal 4 — content series

7-day email sequence.

20 turns.

Goal 5 — data analysis

Quarterly analytics deep-dive.

30 turns.

All ran via single /goal command.

🚀 Want hands-on goals coaching? AI Profit Boardroom has weekly live coaching to wire goals into your workflow. → Join here

Goals Vs Manual Prompting

Quick comparison.

Aspect	Manual prompting	Hermes goals
Engagement	High (every step)	Low (one command)
Coherence	Drift risk	Judge keeps on track
Time	High	Low
Reliability	Depends on you	Depends on judge model
Multi-day	Hard	Resume command

For multi-step work, goals win.

Compounding Effect

Daily goal use = hours/day saved.

Week 1: 5 hrs saved.

Month 1: 20-30 hrs saved.

Quarter 1: 60-90 hrs saved.

That's a huge productivity unlock.

Common Member Questions

Three.

"Is judge model accurate?"

Generally yes.

For nuanced goals, use a stronger judge model.

"Can goals run overnight?"

Yes — until budget exhausted or completed.

"What if goal stalls?"

Goal status will show.

Clear and re-prompt with specifics.

Installing Today

Three steps.

Step 1 — `hermes update`

Pulls goals feature.

Step 2 — Try `/goal` with a small task

Like "build me a markdown file with 5 ideas on X."

Step 3 — Iterate

Try bigger goals.

Adjust config.

Build the muscle.

When To Use Goals Vs Skills

Both have place.

Goals

Open-ended multi-step work.

Skills

Repeated patterns.

Combine: skill within goal.

Quality Of Output

Honest assessment.

Strengths

Persistence.
Multi-step coherence.
Judge keeps on track.

Weaknesses

Can drift on ambiguous goals.
Big goals need careful framing.

For most pro use cases, output quality is excellent.

FAQ — Hermes Agent Goals

Is it really free?

Yes — Hermes is open source.

How to update?

hermes update in terminal.

Best first goal?

Small task — 5-10 turns.

Can goals fail?

Yes — turn budget exhaust.

Cost to run locally?

Free with Ollama.

Best paired feature?

Memory + multi-agent.

Worth it for solo work?

Yes — eliminates micro-prompting.