# Kejue documentation > Automation-first documentation for the reviewed Kejue Voice and Kejue Inference production APIs. This index is generated from the complete documentation navigation. Dashboard-only routes, provider callbacks, and operator diagnostics are intentionally outside the public API. ## Machine-readable contract - [OpenAPI JSON](/api-reference/openapi.json): Deterministic OpenAPI 3.1 contract for generators, CLIs, MCP servers, and agents. Covers both products; Kejue Inference owns every path under /inference. - [Canonical OpenAPI YAML](/api-reference/openapi.yaml): The exact source contract rendered by the API reference. - [Kejue Voice API reference](/voice/api-reference): Human-readable operation, schema, response, and error reference. - [Kejue Inference API reference](/inference/api-reference): Human-readable reference for the speech model endpoints. ## Kejue Voice — Get started - [Welcome to Kejue Voice](/voice): Build voice-AI agents that answer phones, make calls, and talk in the browser — with the guardrails regulated enterprises need. - [How Kejue Voice works](/voice/concepts/how-sonus-works): The five-minute mental model: agents, versions, calls, pipelines, and what happens after a conversation ends. - [Quickstart](/voice/quickstart): Get an API key, create your first agent, and know how to read an error — in about five minutes. - [Web quickstart](/voice/quickstarts/web): Talk to your agent in the browser — mint credentials on your backend, connect with the SDK. - [Phone quickstart](/voice/quickstarts/phone): Connect a SIP trunk or Twilio, attach a phone number to your agent, and dial in. ## Kejue Voice — AI assistants - [Use Kejue Voice from Claude (MCP)](/voice/ai-assistants/mcp): Connect Claude or any MCP client to Kejue Voice with one URL, sign in, and build, test and review voice agents across your teams in plain language. ## Kejue Voice — Build agents - [Agents](/voice/agents/overview): How the draft → version → deploy lifecycle keeps every agent change reviewable and reversible. - [Editing the draft](/voice/agents/settings): How PATCH /v1/agents/{id}/draft works — merge-patch semantics, how to clear a value, size limits, and where every settings field is documented. - [Speech & turn-taking](/voice/agents/speech): Pipeline mode, models and voice, transcription hints, end-of-turn detection, and background noise. - [Conversation behavior](/voice/agents/behavior): Opening line, temperature, call limits, inactivity handling, contact memory, and resuming dropped calls. - [Transfers & call actions](/voice/agents/call-actions): Transfers, agent handoff, scheduled callbacks, waiting for the caller, end call, tool and MCP assignments — and following or cancelling transfer attempts at runtime. - [Prompt variables](/voice/agents/prompting): Call-specific context in prompts and tools via {{scope.name}} templates — no new agent version per customer. - [Guardrails](/voice/agents/guardrails): Fail-closed response-safety policies: author reusable rule sets, attach them to agents, and nothing unsafe gets spoken. - [Pronunciation](/voice/agents/pronunciation): Teach agents to say brands, acronyms, and codes correctly with ordered text-replacement tables applied just before TTS. - [Voices & models](/voice/agents/voices): Discover the enabled models and voices your workspace can use, and plug their IDs into agent settings. ## Kejue Voice — Flows - [Flows](/voice/flows/overview): Join the agents you already have into a named graph, and let one call move between them without ever starting over. - [Handoffs](/voice/flows/handoffs): Edges carry a plain-language condition, replace the agent's own handoff configuration for the call, and leave one transcript behind. - [Pointing calls at a flow](/voice/flows/targeting): An inbound number, a single outbound call, or a test call can target a flow instead of an agent — the start agent answers. - [After a flow call](/voice/flows/after-call): A flow carries its own summary, extraction, and webhook configuration — and it replaces every member agent's for the calls it runs. ## Kejue Voice — Groups - [Groups](/voice/groups/overview): Split live calls across agents you already have, by weight, and compare how each one performs. - [Compare agents with groups](/voice/groups/compare-agents): Run two or more agents side by side on live calls, read who is ahead, and move traffic to the winner. ## Kejue Voice — Phone - [SIP trunks](/voice/telephony/sip-trunks): Connect your own carrier or a Twilio account so your agents can answer and place phone calls. - [Phone numbers](/voice/telephony/phone-numbers): Put an E.164 number on a trunk, route its calls to an agent, and control how it dials out. ## Kejue Voice — Web - [Web calls](/voice/web/overview): Run your agent in the browser — voice over WebRTC or text chat (beta). Your backend mints a short-lived token; media flows browser-to-Kejue Voice directly. - [Voice calls with the SDK](/voice/web/voice-sdk): Build a browser voice-call UI with KejueWebCall — states, transcripts, tool activity, mute, and audio handling. - [Text chat with the SDK](/voice/web/chat-sdk): Build a chat UI with KejueTextCall — send messages, stream replies with deltas, and handle interruptions. - [SDK reference](/voice/web/sdk-reference): Every export of @kejue/voice-web — startCall, KejueWebCall, KejueTextCall, options, events, and types. ## Kejue Voice — Tools & integrations - [HTTP tools](/voice/tools/http-tools): Give your agent outbound API calls — reusable request templates it fills in and fires mid-conversation. - [MCP servers](/voice/tools/mcp-servers): Connect a remote MCP server, review the tools it exposes, and attach the approved set to your agents. - [Secrets](/voice/tools/secrets): Store API tokens and credentials for your tools — encrypted, write-only, referenced by name. ## Kejue Voice — Campaigns & outbound - [Campaigns](/voice/outbound/campaigns): Upload a recipient list, pick an agent and caller ID, and we dial through it inside your calling policy. - [Outbound calls](/voice/outbound/jobs): Place a single durable outbound call: schedule it, respect contact hours, retry sensibly, and never dial twice. - [Outbound policy & caller ID](/voice/outbound/policy): The organization-wide rules for outbound dialing: contact hours, retry behavior, caller-ID number pools, and capacity. - [DNCR screening](/voice/outbound/dncr): Check UAE destinations against the Etisalat Do Not Call Registry before we dial them. - [Contacts](/voice/outbound/contacts): A canonical phone directory: one record per E.164 number, with tags, call history, and cross-call memory. ## Kejue Voice — After the call - [Webhooks](/voice/webhooks/overview): Get a signed HTTP POST to your endpoint when a call finishes or a campaign changes state. - [Webhook event payloads](/voice/webhooks/events): Field-by-field reference for call.completed and the campaign lifecycle events. - [Verifying webhook deliveries](/voice/webhooks/security): Check the HMAC signature, handle retries and duplicates, and rotate signing secrets safely. - [Calls](/voice/analytics/calls): Every conversation becomes a call record — list it, read it redacted, act on it live, and control how long it lives. - [Extraction objects](/voice/analytics/extraction-objects): Reusable schemas that turn every finished call into structured data — a policy number, an incident type, a boolean. - [Analysis boards](/voice/analytics/boards): Save aggregate queries over calls, extractions, guardrail triggers, tool calls, and contacts — preview first, then chart and export. - [Usage and metering](/voice/analytics/usage): Aggregate workspace usage over time, or stream the per-event inference ledger — the same data your bill is built from. - [Alerts](/voice/analytics/alerts): Monitor tools, webhooks, extractions, outbound calls, SIP trunks, campaigns and DNCR checks with Team-scoped rules, incident history and email notifications. - [Platform monitoring](/voice/analytics/platform-monitoring): Use Platform health for operational failures and Performance for call response and provider request timing. - [Human-handled calls](/voice/analytics/human-calls): Ingest recordings or transcripts of calls your human agents handled, and analyze them on the same rubric as AI calls. ## Kejue Voice — Testing & evals - [Evals](/voice/evals/overview): Test your agent like software — simulated callers run your scenarios, reusable checks grade every conversation. - [Eval checks](/voice/evals/checks): Reusable graders — LLM judges, tool-trace rules, and conversation-flow facts — you attach to suites with weight and blocking semantics. - [Running evals](/voice/evals/running): Execute a suite against an agent version — simulated, with live tools, or as grading of real production calls. ## Kejue Voice — Trust & governance - [Deployment & data residency](/voice/trust/deployment): Where Kejue Voice runs, where your call data lives, and the controls — encryption, redaction, isolation, audit — that stand around it. - [Compliance](/voice/governance/compliance): The audit log, PII masking and its dictionary, and the break-glass workflow for un-redacted call data. - [Roles, members, and API keys](/voice/governance/roles-and-members): One permission vocabulary for humans and API keys — discover it, build roles from it, and manage who holds what. - [Workspace settings](/voice/governance/workspace): Who am I, and what are this workspace's defaults — identity discovery plus the mutable settings document. ## Kejue Voice — API essentials - [Authentication & permissions](/voice/essentials/authentication): Bearer API keys, the resource:action permission model, and how delegation stays least-privilege. - [Errors](/voice/essentials/errors): One envelope for every failure, and a retry strategy you can implement once. - [Error codes](/voice/essentials/error-codes): Every stable error code the Kejue Voice API can return, with its HTTP status and whether a retry can help. - [Rate limits](/voice/essentials/rate-limits): 600 requests per key per minute by default — here's how to read the headers and back off. - [Idempotency](/voice/essentials/idempotency): Why every POST takes an Idempotency-Key, and how retries can never create duplicates. - [ETags & conditional writes](/voice/essentials/conditional-requests): Read, get an ETag, write with If-Match — and never overwrite someone else's change. - [Pagination, filters & search](/voice/essentials/pagination): Cursor-based paging: follow next_cursor, keep your filters stable, stop at null. - [Asynchronous workflows](/voice/essentials/async-operations): Some operations commit intent now and finish later — here's the one polling loop that handles all of them. - [Logging (on-prem)](/voice/essentials/logging-onprem): Structured logs, redaction, and correlation for self-hosted deployments. ## Kejue Voice — API reference - [API overview](/voice/api-reference): Base URL, authentication, and how this reference is organized — read this before your first request. ## Kejue Voice — Changelog - [Changelog](/voice/changelog): Kejue Voice releases and the capabilities each one delivered, including historical Sonus releases. ## Kejue Inference — Get started - [Kejue Inference](/inference): Generate speech and transcribe audio with published Kejue model names. - [Access and API keys](/inference/access): Workspace entitlement, Inference-only credentials, and precise permissions. ## Kejue Inference — Speech models - [Text to speech](/inference/text-to-speech): Generate streamed PCM audio over HTTP or a live WebSocket, with stable model names and approved voices. - [Speech to text](/inference/speech-to-text): Transcribe bounded WAV files or stream PCM audio for final segment transcripts. ## Kejue Inference — Billing - [Minutes, limits, and monitoring](/inference/usage): Separate speech wallets, concurrency, reservations, and latency trends. ## Kejue Inference — API reference - [API overview](/inference/api-reference): Base URL, Inference keys, and how this reference is organized. ## Kejue Inference — Changelog - [Changelog](/inference/changelog): Kejue Inference releases and the capabilities each one delivered.