@sqlrooms/ai / AiSliceState
Type Alias: AiSliceState
AiSliceState =
object
Properties
ai
ai:
object
| Name | Type | Description |
|---|---|---|
initialize()? | () => Promise<void> | - |
destroy()? | () => Promise<void> | - |
config | AiSliceConfig | - |
promptSuggestionsVisible | boolean | - |
draftPrompt | string | Transient composer prompt used before the first session is created. |
apiKeyErrors | Record<string, boolean> | Tracks API key errors per provider (e.g., 401/403 responses) |
tools | StoredToolSet | - |
toolRenderers | ToolRendererRegistry | - |
remoteClientToolNames | string[] | Executable local tools that await browser output with remote chat. |
timeouts | AiTimeoutOptions | Opt-in timeout policy for chat runs and tool execution. |
getProviderOptions? | GetProviderOptions | - |
setConfig() | (config) => void | - |
setPromptSuggestionsVisible() | (visible) => void | - |
setDraftPrompt() | (prompt) => void | Update the transient composer prompt used when no session is active. |
setApiKeyError() | (provider, hasError) => void | Set API key error flag for a provider |
hasApiKeyError() | () => boolean | Check if there's an API key error for the current provider |
getAbortController() | (sessionId) => AbortController | undefined | - |
setAbortController() | (sessionId, controller) => void | - |
getSessionChat() | (sessionId) => Chat<UIMessage> | undefined | Return the ephemeral AI SDK chat for a session. |
setToolCallSession() | (toolCallId, sessionId) => void | Map toolCallId -> sessionId for long-running tool streams (e.g. agent tools) |
getToolCallSession() | (toolCallId) => string | undefined | - |
agentProgress | Record<string, AgentToolCall[]> | Live progress for sub-agent tool calls, keyed by parent toolCallId |
updateAgentProgress() | (parentToolCallId, toolCalls) => void | - |
clearAgentProgress() | (parentToolCallId) => void | - |
devtools | AiDevtoolsState | Devtools-only agent snapshot state and controls. |
pendingSubAgentApprovals | Record<string, PendingSubAgentApproval> | Pending approval requests from sub-agent tools with needsApproval |
requestSubAgentApproval() | (approval) => void | - |
resolveSubAgentApproval() | (approvalId, approved) => void | - |
clearSubAgentApproval() | (approvalId) => void | - |
writeAbortSnapshot() | (toolCallId, snapshot) => void | Transient abort snapshots for nested agent progress propagation |
readAbortSnapshot() | (toolCallId) => AgentProgressSnapshot | undefined | - |
clearAbortSnapshots() | () => void | - |
isSummarizing | boolean | True while "summarize and continue" is in progress |
setIsSummarizing() | (value) => void | - |
toolTimings | Record<string, ToolTimingEntry> | Per-tool-call timing entries, keyed by toolCallId |
setToolTiming() | (toolCallId, entry) => void | - |
getToolTimings() | () => Record<string, ToolTimingEntry> | - |
setPrompt() | (sessionId, prompt) => void | - |
getPrompt() | (sessionId) => string | - |
setIsRunning() | (sessionId, isRunning) => void | - |
getIsRunning() | (sessionId) => boolean | - |
addAnalysisResult() | (message) => void | - |
sendPrompt() | (prompt, options?) => Promise<string> | - |
startAnalysis() | (sessionId, attachments?) => Promise<void> | - |
startAnalysisWhenReady() | (sessionId, attachments?) => Promise<boolean> | Compatibility entry point; session controllers are ready synchronously. |
startNewSession() | (name, prompt) => Promise<void> | - |
cancelAnalysis() | (sessionId) => void | - |
setAiModel() | (modelProvider, model) => void | - |
getSelectedModel() | () => ModelSelection | Resolve the model/provider that would be used right now: the current session's selection when a session exists, otherwise the default that a lazily created session would receive. Useful before any session exists, e.g. to know which provider a first-time API key belongs to. |
hasResolvableModel() | () => boolean | Whether a model is resolvable by any configured path: a custom-model factory supplied to AiSliceOptions.getCustomModel, or — once a session exists — its provider/model pair being present in the @sqlrooms/ai-settings model list. With no session yet, the resolved default is assumed available. Only checks that a factory was configured; never calls it, since invoking it may have side effects and a factory returning undefined means "configured but not currently ready". Prefer this over re-deriving readiness from aiSettings.config, so UI and runtime agree on one source of truth. |
requiresApiKey() | () => boolean | Whether the model resolution path in effect needs a browser-held API key. false only when a chatEndPoint is configured (the request is sent server-side), or when AiSliceOptions.getCustomModel is configured and currently returns a model, which carries its own credentials. A factory returning undefined still needs one: the transport then falls back to the built-in OpenAI-compatible client. Unlike AiSliceState.ai.hasResolvableModel this invokes the factory, so keep it cheap — it is called during render. The result is cached per selection and retired when the AI settings change. |
createSession() | (name?, modelProvider?, model?) => string | Create a new chat session, make it the current session, and open it in a tab. When modelProvider/model are omitted the current selection (or configured defaults) are used. |
forkSessionFromMessage() | (args) => string | undefined | - |
getSessionForkOrigin() | (sessionId) => AiSessionForkOrigin | undefined | - |
switchSession() | (sessionId) => void | - |
resetCurrentSession() | () => void | Clear the current session selection (sets currentSessionId to undefined) without deleting any session, returning the UI to the start/new-chat state. A fresh session is created lazily on the next message. |
renameSession() | (sessionId, name) => void | - |
deleteSession() | (sessionId) => void | - |
setOpenSessionTabs() | (tabs) => void | - |
togglePinSession() | (sessionId) => void | Toggle the pinned state of a session. Pinning an unknown session id is a no-op; unpinning is always allowed (also used to drop stale ids). |
isPinnedSession() | (sessionId) => boolean | |
getCurrentSession() | () => ChatSessionSchema | undefined | - |
getSessionRunContext() | (sessionId) => AiRunContext | undefined | - |
setSessionRunContext() | (sessionId, runContext) => void | - |
getSessionDraftContextItemIds() | (sessionId) => string[] | undefined | - |
setSessionDraftContextItemIds() | (sessionId, itemIds) => void | - |
setSessionUiMessages() | (sessionId, uiMessages) => boolean | - |
persistTimedOutSession() | (sessionId, uiMessages, timeoutMessage) => void | Persist a terminal timeout result and force the chat runtime to reload. |
getAnalysisResults() | () => AnalysisResultSchema[] | undefined | - |
deleteAnalysisResult() | (sessionId, resultId) => void | - |
getAssistantMessageParts() | (analysisResultId) => UIMessage["parts"] | - |
findToolRenderer() | (toolName) => ToolRenderer | undefined | - |
getApiKeyFromSettings() | (provider?, model?) => string | Resolve the API key for the outbound provider. When provider/model are omitted the current session's provider (or the default) is used; callers targeting a specific provider (e.g. one-shot sendPrompt) must pass it so the key matches the endpoint the request is sent to. |
getBaseUrlFromSettings() | (provider?, model?) => string | undefined | Resolve the base URL for the outbound provider. See AiSliceState.ai.getApiKeyFromSettings for the provider/model override semantics. |
getMaxStepsFromSettings() | () => number | - |
getFullInstructions() | (sessionId?) => string | - |
getLocalChatTransport() | (sessionId) => DefaultChatTransport<UIMessage> | - |
chatEndPoint | string | Optional remote endpoint to use for chat; if empty, local transport is used |
chatHeaders | Record<string, string> | - |
getRemoteChatTransport() | (sessionId, endpoint, headers?) => DefaultChatTransport<UIMessage> | - |
onChatFinish() | (args) => void | - |
onChatError() | (sessionId, error, messages?) => void | - |
