Agent configuration reference
Current custom-agent fields, model controls, Tools and Connections states, and Ask Salt activity indicators.
This reference catalogs the current custom-agent fields and model settings, the states shown by Tools and Connections, and the tool activity indicators shown in Ask Salt.
Custom-agent fields
| Field | New-agent default | Description |
|---|---|---|
| Name | New agent | Display name for the custom agent. The maximum length is 160 characters. |
| Description | Empty; shown as No description yet. | Short summary shown with the agent. |
| Icon | Bot icon | Lucide icon name or uploaded SVG used for the agent. Salt sanitizes uploaded SVG content and limits it to 4,096 characters. |
| Color | No color | Optional accent: Blue, Green, Amber, Red, Purple, or Mint. |
| Model | Claude Sonnet 5 from anthropic | Model used by the agent. The menu contains the platform's currently enabled models, grouped by provider. |
| Behavior | Empty | System prompt that defines how the custom agent responds and orchestrates. |
| Tools | Web Search enabled | Built-in tools and attached Agentic Tools available during a chat. Published Agentic Tools carry the exact Pipeline badge in the picker. |
| Connections | None | Live Connections and knowledge bases available as chat context. |
If a saved model is no longer in the enabled model list, the menu retains it as model-id (unsupported) and shows model-id is not in the supported list. Pick a supported model.
Model controls
The Advanced section contains provider-aware reasoning and sampling controls.
| Control | Availability | Options or editor range | Initial state | Description |
|---|---|---|---|---|
| Reasoning | Anthropic models | Default, Off, Low, Medium, High | Default; no saved override | Reasoning effort. Default uses the platform setting, while Off disables reasoning. |
| Reasoning | OpenAI reasoning models | Default, None, Low, Medium, High, Extra high, Max | Default; no saved override | Reasoning effort. Default uses the provider setting, while None disables reasoning. |
| Temperature | All models | Anthropic: 0–1; other providers: 0–2, in 0.1 increments | Off; 0.7 when enabled | Lower values are more focused and deterministic; higher values are more varied. |
| Top P | All models | 0–1, in 0.05 increments | Off; 1 when enabled | Nucleus sampling cutoff. |
| Max output tokens | All models | 256–32,000, in increments of 256 | Off; 4,096 when enabled | Upper bound on tokens generated by each top-level LLM step. Delegated specialist steps use provider defaults. |
Temperature and Top P remain disabled while Anthropic reasoning is active or OpenAI reasoning is not set to None. Claude 4.7 and later, including Claude 5, use reasoning effort instead of Temperature and Top P, so Salt does not apply those sampling values. On other Anthropic models, Temperature takes precedence when both Temperature and Top P are set.
Tools
Tools contains the built-in Web Search tool and published Agentic Tools available to the current organization. The sub-header shows Tools · n, and Add tools opens the full catalog.
| State | Exact badge or copy | Meaning |
|---|---|---|
| Enabled built-in tool | Built-in | A Salt-provided tool attached to the agent. |
| Enabled Agentic Tool | Pipeline | A published Agentic Tool attached to the agent. |
| Unavailable bound Agentic Tool | Unavailable and This pipeline no longer exists. Remove it from this agent. | The bound Agentic Tool no longer resolves in the viewer's catalog. |
| Built-in catalog item | Built In | A Salt-provided tool in Add tools. Built-in items remain pinned above Agentic Tools. |
| Agentic Tool catalog item | Pipeline | A published Agentic Tool in Add tools. |
| No enabled tools | No tools enabled. Choose Add tools to browse the catalog. | The agent has no tool bindings. |
| Catalog loading | Loading tools… | The tool catalog is loading and has no rows to show. |
| No search results | No tools match “query”. | No catalog item matches the current search. |
The Add tools catalog includes Search tools…, the A–Z and Enabled first sort options, an n enabled count, and Showing x of y. Agentic Tool details can include a description, When not to use, and up to eight Inputs.
Connections
Connections is the agent editor's label for chat data sources; it does not mean an Edge on the Pipeline canvas. The sub-header shows Connections · n, and Add connections opens available live Connections and knowledge bases.
| State | Exact badge or copy | Meaning |
|---|---|---|
| Attached live Connection | Connection | A live data source attached to the agent. |
| Attached knowledge base | Knowledge base | An indexed knowledge base attached to the agent. |
| Unavailable bound source | Unavailable and This source is no longer available. Remove it from this agent. | The attached source no longer resolves in the viewer's catalog. |
| No attached Connections | No connections attached. Choose Add connections to browse available data sources. | The agent has no attached sources. |
| Catalog loading | Loading connections… | The source catalog is loading and has no rows to show. |
| No search results | No connections match “query”. | No available source matches the current search. |
The Add connections catalog includes Search connections…, the A–Z and Enabled first sort options, an n enabled count, and Showing x of y.
Ask Salt activity indicators
Ask Salt renders a tool's purpose with an activity icon, check, X, or dot. It does not display internal tool-call status labels or raw returned data in the activity row. See the Ask Salt reference for the indicator mapping.