Skip to main content
TripleBits Apprentice - Software for Agents

Menu

Sign In Register

Create Your First Agent

Create your first Apprentice agent with a focused role, one verified provider and model, limited permissions, optional guardrails, and a first test run.

25 views

Create Your First Agent

After Docker is installed and at least one AI provider is connected, you can create your first Apprentice agent.

For the first agent, keep the scope small. Use one provider, one model, one clear job, and limited permissions. You can expand the agent later after you see how it behaves.

Before You Start

Make sure you have:

  • Docker installed and running.
  • At least one AI provider enabled in Settings > AI Integration.
  • A model selected for that provider.
  • Optional: a folder or project you want the agent to work with.

If Apprentice says "Connect an AI first", open AI Integrations and finish setting up a provider before creating the agent.

Open The Agent Wizard

In Apprentice, select Create Agent from the Dashboard, Agents page, Tasks page, or sidebar.

The agent wizard walks through:

  1. Template
  2. Identity
  3. AI Model
  4. Instructions
  5. Permissions
  6. AI Guardrails
  7. Review

Choose A Starting Point

Start with a template or choose the scratch option.

Templates pre-fill parts of the agent configuration. Starting from scratch gives you a blank agent.

For your first agent, either option is fine. If you are unsure, start from scratch and create a simple project assistant.

Set The Agent Identity

Give the agent:

  • A short name.
  • A description.
  • An icon and color.
  • A personality preset, if useful.

Example:

Project Notes Assistant

Description:

Helps summarize project notes, answer questions, and suggest next actions for one project folder.

Choose The AI Model

Select:

  • Provider
  • Model
  • Account, if the provider supports accounts
  • Runtime limits
  • Optional budget settings

For the first agent, use the provider and model you already verified.

If you are using a paid API provider, set a small budget or limit while testing. If you are using a local/offline runtime, budget tracking may not apply the same way.

Write Instructions

The instructions tell the agent what it is responsible for.

Keep the first version narrow:

You help me understand and organize one project folder.

Summarize files when asked.
Suggest next actions when helpful.
Ask before making changes.
Do not modify files unless I explicitly request it.
Keep answers concise and practical.

You can also attach knowledge base files during this step. Use this for documents the agent should reference, such as notes, specs, or project background.

Set Permissions

Permissions control what the agent can access and execute.

For a first agent, use Ask for Approval. This makes the agent request permission before operations.

Permission modes include:

  • Ask for Approval
  • Auto-Accept Safe Operations
  • Deny All
  • YOLO / Allow All Operations

Avoid YOLO for your first agent.

If the agent needs access to a folder, add it as a volume. A mounted folder appears inside the agent container under the agent's home directory.

Use read-only access when the agent only needs to inspect files. Use read-write access only when the agent should be allowed to modify files.

Configure AI Guardrails

AI Guardrails are optional.

They can evaluate actions such as command execution, file reads, file writes, and network access before the action runs. Guardrails add extra model calls and latency, so you do not need to enable them for every simple first agent.

For your first agent, either leave guardrails off or enable them only if you want extra review around sensitive actions.

Review And Create

On the Review step, check:

  • Agent name and description.
  • Provider and model.
  • Permission mode.
  • Mounted folders.
  • Budget settings.
  • Guardrails, if enabled.

Then select Create Agent.

Apprentice creates the agent and opens its chat.

Send A First Message

Start with a simple test:

Summarize what you can help me with based on your current configuration.

If you mounted a project folder, try:

Inspect the mounted project folder and summarize the main areas. Do not modify files.

Review What Happened

After the first run, check:

  • Which provider and model were used.
  • Whether permissions were requested.
  • Whether the agent tried to access the expected files.
  • Whether the response matched the instructions.
  • Whether the agent needs more or fewer permissions.

This is the core setup loop: create a small agent, run it once, inspect the result, then adjust the configuration.

Troubleshooting

If you cannot create an agent, make sure an AI provider is enabled and has a model selected.

If no model appears, return to Settings > AI Integration and test the provider connection.

If the agent cannot see files, check that the folder was added as a volume.

If the agent cannot run a command, check the permission mode and command restrictions.

If an API or CLI provider fails during the first message, verify that the provider is still authenticated and reachable.

If you hit an agent limit, check your Apprentice license plan.

Next Step

After creating your first agent, learn how local-first execution works in Apprentice.