You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
description: localize('agentHost.config.multiTurnContextRouting.description',"When enabled, Auto model selection sends prior user messages to the router so it scores the conversation so far instead of the latest message alone."),
markdownDescription: nls.localize('chat.agentHost.copilot.multiTurnContextRouting',"When enabled, Auto model selection in Copilot SDK agent sessions routes on the conversation so far, sending prior user messages to the router instead of scoring the latest message alone."),
markdownDescription: nls.localize('chat.agentHost.copilot.modelCapabilityOverrides',"Per-model capability overrides for Copilot SDK agent sessions, keyed by model id (`*` matches every model; a specific entry wins field-by-field), intended for evaluating models against an existing model's profile. Declare an aliased `family` (for example `claude-opus-4.8`) to route the model to that family's tuned system prompt and tool profile without changing the model id sent to the runtime — so a preview model can be evaluated against a known prompt while still running on its own endpoint — a `reasoningEffort` to pin its effort level, `availableTools`/`excludedTools` to filter its tool set, or `modelCapabilities` to override individual capability limits (e.g. vision support, context window size) passed through to the SDK. All overrides apply when a session launches or resumes. On a mid-session model change, only the new model's `reasoningEffort` is applied; the session keeps its launch-time family, tool filters, and model capabilities. Only affects Copilot agent sessions.\n\n**Note**: This is an advanced setting for experimentation."),
0 commit comments