Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Agent capabilities are opt-in features that extend what an agent can understand, produce, retrieve, execute, observe, or secure. For the runtime, type, conversation, and middleware foundations behind these features, see Agents.
Agent types and model connections
Looking for the agent-type and SDK-selection guidance previously hosted on this page?
- Agent concepts explains application-owned chat-client agents, custom agents, and remote agent types.
- Model providers compares inference providers, conversation-history support, and .NET SDK and endpoint options.
- Agent services covers managed and protocol-backed remote agent runtimes.
Input and output
| Capability | Purpose |
|---|---|
| Multimodal | Send images and other supported content to an agent. |
| Structured outputs | Return values that conform to a schema or application type. |
| Background responses | Continue, poll, and reconnect to long-running responses. |
Context and knowledge
| Capability | Purpose |
|---|---|
| RAG | Ground responses with retrieved application knowledge. |
| Declarative agents | Define supported agents through YAML or JSON. |
| Agent Skills | Discover and progressively load reusable instructions, resources, and scripts. |
Execution and autonomy
| Capability | Purpose |
|---|---|
| Tools | Let agents call functions and provider-hosted capabilities. |
| CodeAct | Let the model write programs that coordinate tools through a managed execution provider. |
| Looping | Re-run an agent until a bounded completion condition is met. |
| Background agents | Delegate work to background agents and retrieve task results. |
| Planning and todos | Track plans, operational todos, dependencies, and completion. |
Operations and trust
| Capability | Purpose |
|---|---|
| Observability | Export traces, metrics, and logs. |
| Evaluation | Measure agent quality, safety, and correctness. |
| Agent Hooks | Apply fail-closed governance controls through a shared interception contract. |
| Agent Security with FIDES | Enforce information-flow controls across agent data and tools. |
The Harness Agent assembles many of these capabilities into an opinionated operational agent.