Agent Harness lets you build, configure, and deploy purpose-built AI Agents without writing code. Instead of using one generic assistant for every task, Harness lets you create AI teammates designed around a specific role, workflow, knowledge base, and set of tools. Every team works differently. Support needs an Agent that understands your product and resolves customer questions. Operations needs an Agent that can work with internal systems and processes. Engineering needs an Agent that can investigate issues using technical tools. With Agent Harness, you configure each Agent for the job it needs to do.

​
Core Capabilities

Identity

Define who the Agent is, what its role is, what it should do, and how it should communicate.

Memory

Give the Agent persistent context so important instructions and information carry across conversations.

Skills

Attach reusable instructions, workflows, and response formats. Build a skill once, reuse it across Agents.

Sub-Agents

Create specialist Agents and let a primary Agent delegate work to them when a task needs deeper expertise.

Tools

Connect Agents to real-world Actions so they can search, execute, analyze, and act.

Knowledge Base

Connect your documents and internal knowledge so Agents answer using your organization’s actual information.

​
Configuring an Agent

1

Define Its Identity

Start with a simple description of the Agent’s role, responsibilities, behavior, and communication style. Tell it what it’s there to accomplish.
2

Give It Memory

Add persistent context and instructions that remain available across conversations, so the Agent doesn’t start from zero every time.
3

Add Skills

Attach reusable skills that define how the Agent performs specific types of work.
4

Connect Knowledge

Add your team’s documents and knowledge sources so the Agent can retrieve relevant information when answering questions, keeping responses grounded in your actual business context.
5

Connect Tools

Give the Agent access to the systems it needs to do its job. Tools can be configured with approval controls, so actions that require human judgment never happen automatically.
6

Add Specialist Agents

When a task gets more complex, let your primary Agent delegate work to specialized sub-agents while it coordinates and brings the results together.

​
Approval Gates

For sensitive or consequential actions, a Tool can be configured to require human approval. The Agent can prepare the action, but you decide whether it happens:
This gives you the productivity of AI Agents without giving up control over important decisions. See Actions & MCP for how Tools are configured.

​
Building Agent Teams

Complex work rarely belongs to one specialist. A primary Agent can coordinate a task and delegate specific pieces of work to specialist sub-agents, for example:
  • Research Agent — finds and analyzes relevant information.
  • Data Agent — processes and interprets structured data.
  • Compliance Agent — checks the output against defined policies.
  • Writer Agent — turns the final findings into a structured report.
The primary Agent brings the work together into a single response, instead of one Agent trying to do everything.

​
Chat Experience

Agents built with Harness are used through a chat experience designed for work:
  • Streaming Responses — see the response as it’s generated instead of waiting for the full answer.
  • Interactive Clarification — when a request is ambiguous, the Agent can ask for clarification instead of guessing.
  • Edit & Re-Ask — modify your request and quickly regenerate a response with the updated context.
  • Cross-Session Memory — continue conversations with persistent context instead of starting over every time.
  • Visible Agent Activity — see what the Agent is doing while it works, not just the final result.