AbsoluteJS

Assistants And Tools

Assistant, agent, model, tool, RAG, MCP, migration, and conversation-intelligence primitives.

#Assistant Definition

createVoiceAssistant() defines an assistant with id, system prompt, model, tools, guardrails, memory lifecycle, artifacts, and presets.
defineVoiceAssistant() packages adapters, prompts, language strategy, phrase hints, lexicon, route config, and session conversion.
createVoiceAgent() and createVoiceAgentSquad() run model/tool loops and specialist handoff policies.
createVoiceExperiment() and summarizeVoiceAssistantRuns() compare variants and run summaries.
TS
import { createVoiceAssistant, createVoiceRAGTool } from "@absolutejs/voice";

const assistant = createVoiceAssistant({
	id: "support",
	system: "You are a concise support agent.",
	model,
	tools: [createVoiceRAGTool(collection)]
});

#Models And Routing

createOpenAIVoiceAssistantModel(), createAnthropicVoiceAssistantModel(), createGeminiVoiceAssistantModel(), and createJSONVoiceAssistantModel() normalize model providers.
createAIVoiceModel() bridges @absolutejs/ai models into voice assistant models.
Provider router options support fallback modes, policy presets, weighted routing, health signals, and decision traces.

#Tool Catalog

createVoiceEndCallTool() completes a call with an optional farewell and result.
createVoiceTransferCallTool() exposes typed destinations and calls api.transfer().
createVoiceDTMFTool() validates and forwards keypad digits.
createVoiceVoicemailDetectionTool() marks voicemail and can complete the call.
createVoiceApiRequestTool() wraps HTTP calls with buildBody, buildQuery, buildHeaders, parseResponse, and injected fetch.
createVoiceRAGTool() wraps an @absolutejs/rag collection and extractVoiceRAGCitations() pulls source citations.
createVoiceMCPToolset() exposes MCP tools as voice agent tools.

#Pathways And Workflows

VoicePathway defines state-machine style call flows with typed slots, conditions, actions, transitions, tools, metadata, and validation.
Pathway actions can say text, collect slots, call tools, set slots, transfer, or end the call.
The pathway compiler, runtime, generator, visualizer, and slot collector support structured conversation flows similar to hosted conversational-pathway builders while staying in code.
IVR plans, DTMF tools, outcome recipes, workflow contracts, and tool contracts make deterministic call flows testable alongside LLM-driven assistants.

#Conversation Intelligence

Post-call analysis, mid-call summaries, transcript annotation, semantic turn detection, call disposition, post-call surveys.
Call scorecards, AI scorecards, LLM judges, calibration, quality drift detection, agent performance reports.
Caller memory, assistant memory, CRM linking, CRM contract registry, CRM call logger, variable analytics, cost accounting, and cost prediction.

#Vapi Migration

fromVapiAssistantConfig() maps Vapi assistant JSON into assistant, tools, routeHints, and unsupported migration notes.
Vapi-like tools map to endCall, transferCall, voicemail, apiRequest/function server tools, query/RAG, and DTMF with a caller-supplied factory.
Unsupported migration fields are reported with concrete manual-review guidance.