Deep Agent keeps application infrastructure behind small, replaceable contracts. Choose one extension surface for each responsibility rather than making model prompts carry application secrets or runtime policy.
| Need | Start here |
|---|---|
| Long-lived agent facts and preferences | Persistent memory |
| Reusable, on-demand task instructions | Skills |
| Parallel or specialized delegated work | Subagents and delegation |
| Tools and read-only resources from external systems | MCP and OAuth |
| Deliberate JavaScript evaluation and narrow callable capabilities | Interpreter tools |
DeepHarnessProfile selects the built-ins suitable for a particular model or deployment: workspace, todos, skills, memory, sandbox, interpreter, subagents, and MCP resources. Application-supplied ChatTools remain available regardless of the profile. Use DeepHarnessProfile.defaults() to enable every built-in, or construct a profile with only the capabilities your deployment supports.