AgentsAgents and context

Agents and context

Learn how durable Agent context, Group context, live Connections, and company knowledge shape an Agent's work.

An Agent can use durable context to understand its role and the work around it.

That context can include the Agent’s own standing guidance, broader Group or workspace context, relevant live systems, and other information the current Giga permissions allow it to retrieve.

Agent context is layered

Giga keeps different kinds of context separate so each one can do a clear job.

Context layerWhat it is for
Agent instructionsHow the Agent should behave and approach its role
Agent contextDurable background specific to the Agent
Group contextShared guidance relevant across a broader team or area
Workspace contextCompany-wide standing context
ConnectionsLive information from external systems
SkillsReusable task-specific instructions and methods
Current ChatTemporary instructions and Chat context for the immediate work

The live workspace determines which layers are actually available in a given session.

What belongs in Agent context

Use durable Agent context for background that should continue to matter across repeated work.

Good examples include:

  • the domain the Agent works in
  • stable company terminology relevant to its role
  • durable handoff information
  • recurring background the Agent needs to interpret requests correctly
  • context that is specific to the Agent rather than the entire company or Group

Keep temporary directions in the current Chat.

Context vs instructions

Agent instructions tell the Agent how to work.

Agent context tells the Agent what it should know.

For example:

InstructionsContext
“Separate confirmed facts from inference.”“Our enterprise plan is called Scale.”
“Ask before taking an outward-facing action.”“The Sales team uses MEDDICC terminology.”
“Return concise account summaries.”“North America is the current expansion focus.”

Keeping these separate makes the Agent easier to maintain.

Context vs Track Context Notes

Context Notes are the user-facing durable standing context for a Track.

Agent context belongs with the Agent.

Use Track Context Notes when the information belongs to one ongoing body of work. Use Agent context when the information should travel with the Agent across the work it handles.

For example:

Group and workspace context

An Agent can also operate inside broader company context.

A Group can provide shared instructions or context relevant to the team or area the Agent belongs to.

Workspace-level context can carry company-wide standing guidance.

This lets company-wide rules stay broad, team-specific context stay with the Group, and Agent-specific background stay with the Agent.

Connections provide live context

Durable context is only part of what an Agent can work with.

Connections let an Agent retrieve current information from external systems when its permissions allow it.

That means an Agent can combine:

  • durable background
  • current Chat instructions
  • approved Skills
  • live external data

The exact mix is resolved from the current Giga environment.

Learn about giving Agents access to Connections →

Skills provide reusable methods

A Skill carries reusable instructions for a specific kind of task.

Agent context should hold background the Agent needs to know. A Skill should hold a method the Agent may need to apply.

For example:

  • Agent context: “Our ICP is US-based B2B SaaS with 50–500 employees.”
  • Skill: “How to research an account and return a qualification summary.”

This keeps reusable task methods separate from persistent Agent knowledge.

Permissions still shape retrieval

An Agent should only retrieve company context it is allowed to access.

Its permission boundary can be affected by:

  • the Group it belongs to
  • available shared resources
  • Connections in scope
  • current user permissions
  • the live Giga surface

Company Brain remains permission-aware

Information existing somewhere in the workspace does not automatically make it available to every Agent.

Updating Agent context

The current context tools treat an Agent’s durable standing context as stored context that can be read and replaced.

Giga should read the current Agent context before making a meaningful update, then amend it rather than wiping useful background accidentally.

Meaningful changes should be confirmed before saving.

A simple example

Imagine a Recruiting Agent.

Each layer has a different job, which keeps the Agent’s knowledge easier to understand and maintain.

Image placeholder

Layered diagram showing Workspace context → Group context → Agent context/instructions, with Skills, Chat, and Connections feeding the Agent at runtime.