Your first Agent

What is this

Your agent is an artificial intelligence assistant that runs on your machine. Unlike ChatGPT or other online services, this agent:

• Works with your files and documents
• Does not send your data to the internet
• Is always available (no browser needed)
• Can execute tasks step by step without you having to supervise every action

Think of it as a collaborator who is always available, can read your documents, and executes tasks on your machine — but who needs clear instructions on what to do.

How to start the agent

Open the terminal and type:

hermes --tui

The agent starts and is ready to receive instructions. You can close the terminal at any time — the session is saved automatically.

To return to the last session:

hermes --tui --continue

To start a new session (clear everything and start over):

hermes --tui --new

How to talk to the agent

Write in natural language, as if you were explaining a task to a collaborator.

Examples of what you can ask:

• "Read the file contract-2024.pdf and make a summary of the main points."
• "You have access to the documents/ folder. List all the PDF files there."
• "Write an email to client John Smith confirming tomorrow's meeting."
• "Analyze the financial-report.txt and identify the three largest costs."
• "Organise this task list by priority and add estimated deadlines."

> Important tip: If you want the agent to work with a specific file, tell it the full path. If you want it to work with all files in a folder, tell it which folder.

Complex tasks: the /goal command

For tasks that require multiple steps, use the /goal command. The agent works autonomously until it completes the goal.

Example:

/goal "Read all PDF files in the contracts/ folder, extract the due dates, and create a list sorted by date."

The agent starts working immediately. You can check the progress with:

/goal status

To pause the work (without deleting the goal):

/goal pause

To resume:

/goal resume

To delete the goal completely:

/goal clear

When to use /goal:

• Tasks with more than 2 or 3 steps
• Work that requires reading multiple files
• Repetitive tasks that the agent can do on its own

When NOT to use /goal:

• A simple question
• A single task (e.g. "summarise this file")
• When you want to supervise every step

Security: permissions before acting

Before executing actions that modify files or run system commands, the agent asks for your authorisation. You will see something like:

Approval required
Command: write_file
Path: /home/user/new-document.txt
Allow? (y/n):

This protects you from unwanted changes. Answer y (yes) or n (no).

> General rule: Approve when you understand what the agent is going to do and you agree. Deny when you are unsure or when the action seems wrong.

The agent never deletes files without asking. Never sends data to the internet without asking. Never modifies system files without asking.

Useful commands

These are the commands you will probably use:

Command	What it does
/help	Shows the list of available commands
/new	Starts a new session (clears everything)
/save	Saves the conversation to a file
/continue	Returns to the last session
/goal "text"	Sets a goal for autonomous work
/goal status	Checks the goal's progress
/goal pause	Pauses the goal work
/goal resume	Resumes the goal work
/goal clear	Deletes the goal
/quit	Exits the agent

What the agent does well

• Reading and summarising documents — understands the content and extracts the essentials
• Writing and editing text — drafts emails, reports, documents
• Organising information — classifies, sorts, structures data
• Answering questions — about the files it has access to
• Executing step‑by‑step tasks — when using /goal
• Working 24 hours — as long as the machine is on

What the agent DOES NOT do (yet)

To be honest about what you can expect:

• It is not connected to your email — cannot read or send emails
• It is not connected to your website — cannot update web pages
• It is not connected to your other systems — (billing software, CRM, etc.)
• It does not remember previous sessions — each session starts "clean" (unless you use /continue)
• It does not guess what you want — it needs clear instructions

These capabilities are the second phase. When you are ready to move forward, we can connect the agent to your systems.

Common problems

The agent does not respond:

• Check that the machine is on and has internet (for the AI API)
• Try Ctrl+C and start again with hermes --tui

The agent says it cannot access a file:

• Check the file path (it must be full: /home/user/folder/file.txt)
• Check that you have permissions to read that file

The agent does something unexpected:

• Answer n to the next permission prompt
• Use /new to start a clean session
• Contact us if the problem persists

The session has become very long and the agent seems confused:

• Use /new to start a new session
• Or use /save to save what you need and then /new

When to contact us

Contact us when:

• You cannot start the agent
• The agent is behaving unexpectedly
• You want to add capabilities (connect to systems, add specialised profiles)
• You have questions about how to use a feature
• You want to move forward to the second phase (connection to local systems)

> Final tip: The agent improves with clear instructions. Instead of "help me with this", say "read file X, extract data Y, and create a table with the results organised by Z". The more specific you are, the better the result will be. > > If something does not work as expected, do not get frustrated — start a new session with /new and try rephrasing the instruction in a different way.

---

Support: contact@sophia.host Technical information: #main — irc.sophia.host Web: sophia.host