Agents

Agents is the configuration side of agent work: everything a session draws on, set up before the session starts. These pages mirror the /ai/* product surfaces — agents, environments, models, tools, skills, plugins, MCP, memory, providers, knowledge, and the runs record. Agent Ops is where sessions happen; this section is what those sessions run with. Configure here, run there.

The inventory

Agents list: the configured agent inventory with type, model, runs, status, and owner.
Agent config: profile, runtime policy, and where tools, MCP servers, skills, and knowledge attach.
When to build an agent: the cases that justify an agent, and the ones that do not. Read before configuring anything.
Subagents: scoped helpers owned by an agent config.

Execution targets and models

Environments: context, tools, and policy bundles sessions inherit.
Model registry: which models agent work runs on.
AI providers: provider wiring behind the gateway key.
Gateways: the provider front door for model inference.
AI config: workspace-level model configuration.

Capabilities

Tool registry: atomic capabilities, connector-backed where external.
Toolsets: adoption, call volume, and the no-agent gap.
Catalog: platform-provided tools before tenant install.
Skill registry: repeatable procedures agents follow.
Plugin registry: installable capability packages.
Capabilities: skills, tools, and wiring adopted as one unit.
MCP registry: platform catalog vs. tenant install, OAuth profiles, verify-before-use.

Context and record

Memory: what persists across sessions, and what does not.
Knowledge stores: scoped reference data sessions read.
Secrets: vaulted credentials referenced as grants, never pasted values.
Observability: session records, approval records, and the signals worth tuning on.

What next

Agent Ops: the session runtime this inventory serves.
Starting a session: where agent and environment picks meet.
RBAC in Agent Ops: who may use which agent.