AgentsCreating an Agent

Creating an Agent

Learn how to set up a reusable Agent with the right role, context, access, and Group placement.

Create an Agent when your team needs a named AI role that should behave consistently across repeated work.

A good Agent has a clear purpose, durable instructions, the right context, and only the Connections and Skills it actually needs.

Start with the job the Agent should do

Before creating an Agent, define the role in plain language.

For example:

  • Sales Research Agent that prepares account research
  • Recruiting Agent that helps screen candidates
  • Support Agent that works with customer context
  • Operations Agent that handles recurring operational tasks

Give the Agent a real job

A clear role makes it easier to decide what instructions, context, Connections, and Skills belong with the Agent.

Choose the right Group

Agents operate within Giga’s Group permission model.

Choose the Group based on who should be able to use the Agent and which shared resources should be in scope.

PlacementGood fit
Private GroupAn Agent only you should use
Shared GroupAn Agent intended for a team or shared function

The Group helps determine the audience around the Agent.

Learn about personal vs shared work →

Build the Agent

Name the Agent

Use a clear name that reflects the role people will recognize later.

Define its role

Describe what the Agent is responsible for and the kind of work it should handle.

Add durable instructions

Capture the working rules, constraints, terminology, and expected approach the Agent should follow repeatedly.

Add relevant context

Give the Agent the durable background it needs for its role.

Give it the right Connections

Make approved external systems available when the Agent needs live data or actions.

Add useful Skills

Reuse task-specific instructions instead of packing every method into the Agent itself.

The exact creation controls can vary with the current Giga product surface. The live workspace remains the source of truth for what can be configured.

Keep the role focused

An Agent should have a coherent job.

If the instructions start covering several unrelated roles, consider creating separate Agents or moving reusable task methods into Skills.

For example:

This keeps the Agent readable and makes the Skill reusable elsewhere.

Add instructions that will still matter later

Agent instructions should describe durable behavior.

Good examples include:

  • what the Agent is responsible for
  • which goals matter most
  • terminology it should use
  • recurring constraints
  • how it should structure its output
  • when it should ask for confirmation
  • which kinds of tasks should be escalated or left to a person

One-off requests belong in the current Chat.

Learn about Agent instructions →

Give the Agent only the context it needs

An Agent can use durable context relevant to its role.

That context should stay focused on what the Agent repeatedly needs to understand, rather than becoming a dump of every piece of company information.

The live permission model still controls what the Agent can actually retrieve.

Learn about Agents and context →

Add Connections deliberately

A Connection can give the Agent access to live systems such as a CRM, support tool, database, or MCP server.

Before making a Connection available, check:

  • whether the Agent actually needs it
  • whether the Agent’s Group has the right access scope
  • what actions the external account allows
  • whether the Connection should be shared with that audience

An Agent should not automatically gain every Connection in the workspace.

Learn about giving Agents access to Connections →

Add Skills for reusable methods

Use Skills for methods the Agent may need repeatedly.

For example, a Recruiting Agent could use:

  • a candidate-screening Skill
  • an interview-summary Skill
  • a sourcing-research Skill

This keeps what the Agent is separate from how a specific task should be done.

A simple example

Imagine you are creating a Customer Support Agent.

PartExample
RoleHelp the support team investigate and resolve customer issues
InstructionsBe concise, cite the source of customer facts, confirm before outward-facing actions
ContextSupport terminology and durable team guidance
ConnectionsSupport platform and approved customer systems
SkillsTroubleshooting and response-drafting methods
GroupCustomer Support

The result is a reusable Agent the support team can return to with the same working model each time.

Image placeholder

Creation flow showing Role → Group → Instructions → Context → Connections → Skills → Ready to use.

After creation

Once the Agent exists, you can refine:

Start focused. Add more access or instructions when the Agent’s real work shows that it needs them.