Skip to content
Community open-source Web · 0.2.27 release docs · 0.2.27Content reviewed: 2026-10-23View Markdown

2.1 AI Team, Agent Studio, and Agent Harness ​

Teloa is an AI-Native Team Studio for building, managing, and running teams of AI employees. You set goals, grant access, and review results; employees collaborate on deliverables within those boundaries. Roles, knowledge, skills, tools, memory, and permissions define lasting jobs. Start with the work you need done and build the team as needed.

1. Three product layers ​

HumanSet goals · Define boundaries · Make key decisions
Team and delivery

AI Team

Know who owns the work and how it is progressing

A
AliceContent strategist
B
BenData analyst
C
CaseyProject coordinator
Business Projects Work groups Tasks Automations

Employees keep a lasting role, responsibilities and memory. They collaborate on business and project work to deliver results.

How each employee works

Agent Studio

Configure capabilities and boundaries

RoleIdentity and responsibilities
KnowledgeReferences and facts
SkillsReusable ways of working
Tools / MCPConnected business systems
MemoryRole experience and feedback
PermissionsAction and approval boundaries
Execution foundation

Agent Harness

Run the work and record what happens

ModelsModel
EnvironmentEnvironment
Tool callsTool Gateway
SandboxSandbox
ApprovalApproval
ObservabilityObservability
Task teamA conversation or AI employee can form a temporary task team. Task assistants do not automatically become AI employees.
People set goals and boundaries. The AI team carries the work forward.Product architecture · Currently built primarily on DSH; availability depends on the release.

TELOA = AI Team + Agent Studio + Agent Harness

Product layerPurposeCore concepts
01 · AI TeamBuild and run your AI teamEmployees, tasks, collaboration, business areas, projects, automation
02 · Agent StudioDefine how each AI employee worksRoles, knowledge, skills, tools / MCP, memory, permissions
03 · Agent HarnessSupport safe, continuous, controlled executionModel, Environment, Tool Gateway, Sandbox, Approval, Observability

Team is the product, Agent is the unit of work, and Agent Harness is the foundation that runs the work. These product layers do not require three new pages or services. See Source architecture for the implementation view.

1.1 AI Team: employees and delivery ​

You need to know who your employees are, what they are doing, what needs you, and whether the work meets your expectations. Business areas organize ongoing work; projects connect tasks, employees, work groups, and results around a delivery goal. Tasks and automations manage one-off and recurring work.

AI Team means your lasting business team. A task team is temporary delegation within a task, started by a conversation or employee. Task assistants do not automatically become employees. See People and collaboration.

1.2 Agent Studio: six working elements ​

ConfigurationQuestion
RoleWho are you, and what are you responsible for?
KnowledgeWhat should you know, and which references may you use?
SkillHow do you work, and which methods can you reuse?
Tools / MCPWhich business systems and tools can you use?
MemoryWhich role experience and feedback has sources and should be retained, reused, or withdrawn?
PermissionsHow far may you act, and when is a person needed?

These elements define a lasting employee together; they do not require six configuration pages. Memory comes from work and feedback with sources. See Automatic memory for usage and management boundaries.

MCP is one protocol for connecting external capabilities. Configuring a tool does not authorize every action. Skills provide established methods while leaving employees able to plan within their permissions.

1.3 Agent Harness: reliable execution ​

Agent Harness is the agent engine: it calls models, provides the execution environment, and handles tool calls, sandbox isolation, approvals, and execution visibility. Teloa uses native engine capabilities first and adapts employee identity, business scope, and delivery rules. DSH is the current engine. Support for other engines, enterprise security gateways, and distributed deployment is described in Editions.

1.4 From setup to autonomous work ​

  1. Recruit employees and define their responsibilities.
  2. Add knowledge, business references, and context.
  3. Teach skills so established methods can be reused.
  4. Connect tools or MCP services to real business systems.
  5. Set permissions, including read, write, and approval boundaries.
  6. Start work through tasks, supported events, or schedules while the environment is online.
  7. Bring key decisions back to a person: approvals, exceptions, high-risk actions, and delivery review.

Start with an employee template or AI team template; you do not need to repeat setup for every task. Autonomous work remains subject to permissions, supported triggers, and whether the environment is available.

2. Skills, connections, and extension packages ​

A skill describes how to work. A connection provides external information or tools. A task template supplies reusable task inputs, goals, and delivery requirements. Templates can reference skills, and a skill can serve different tasks.

An extension package is a unit of installation and distribution. It can provide multiple skills, connections, or execution capabilities. These are contents provided by a package. A capability should identify its source package; package management should explain what it provides and which business areas use it.

“Official” and “third-party” describe the publisher. “Built-in” and “user-installed” describe delivery. Neither distinction alone determines whether an item can be disabled or removed. Required system components and optional extensions need different treatment.

3. What an AI team template contains ​

An AI team template can combine seven kinds of content: AI employees, skills, reference knowledge, connections, business boards, task templates, and required apps. It supplies business configuration, rather than representing a single engine extension.

For example, a marketing team template might define content employees, writing skills, product information, a data connection, and a weekly report template. After adding it, check the content, connections, and employee permissions separately. Adding the template does not make all its dependencies ready. See Business resource packages to create one.

4. Installed does not mean available or authorized ​

Fact to checkMeaning
InstallationContent is installed in the relevant environment
EnablementA loading or enablement choice has been saved
RunningThe component has actually loaded, or has an error or requires a restart
Connection and authenticationThe external service, keys and authorizations are usable
Business authorizationA particular employee may perform particular operations within a defined scope

These are independent dimensions, not a mandatory five-step sequence. A skill may need no authentication. An enabled extension that fails to load remains unavailable. A working connection does not grant every employee permission to publish, change, or delete information.

For a weekly marketing report, the task template defines inputs and delivery requirements, skills provide the analysis method, connections supply data, and the employee acts within its permissions. The engine records execution. The task requirements determine whether the result is accepted.

5. Where to manage things ​

Marketplace supports discovery and selection. Marketplace > My apps and skills manages installed apps and skills; business and employee configuration have their own entry points. Connections and environment in Settings directly contains native extension management and run configuration. Component status is for dependencies. Installation, enablement, running state, and business authorization are checked separately.

Settings reuses the official management interface and save operations. Teloa manages business authorization separately. See the capabilities guide and Capability and engine integration boundaries.