TypeScript SDK - Interhuman AI Beta Docs

Classes

AuthClient

Exchanges API-key credentials for a short-lived bearer access token. The token returned by createToken carries only the scopes you request and expires after expires_in seconds (typically 15 minutes). For hands-off token lifecycle management, prefer the top-level InterhumanClient, which uses a TokenManager to fetch and refresh tokens for you.

Constructors

Constructor

new AuthClient(options?): AuthClient;

Parameters

Parameter	Type
`options`	`AuthClientOptions`

Returns

AuthClient

Methods

createToken()

createToken(request): Promise<TokenResponse>;

Mint an access token for the given credentials and scopes.

Parameters

Parameter	Type
`request`	`TokenRequest`

Returns

Promise<TokenResponse>

Throws

when the credentials are invalid (401), lack a requested scope (403), or the request is otherwise rejected.

createClientToken()

createClientToken(request): Promise<ClientTokenResponse>;

Mint a short-lived, capped client token for direct browser use. Call this from your backend with your API key; the returned token is safe to hand to a browser to call the upload, stream, or realtime endpoints directly. It grants the requested scopes (default interhumanai.stream) and carries the caps you set, which the API enforces across upload, stream, and realtime (the video budget spans all three). Omitted fields are left unset (the server applies its defaults, e.g. a single concurrent session and a 300s TTL).

Parameters

Parameter	Type
`request`	`ClientTokenRequest`

Returns

Promise<ClientTokenResponse>

Throws

when the credentials are invalid (401), the key lacks a requested scope (403), or the mint rate limit is exceeded (429).

revokeClientToken()

revokeClientToken(request): Promise<void>;

Revoke a previously minted client token. Call this from your backend with the same API key that minted the token. Once revoked, the token is rejected on new requests and any live streaming session it opened is torn down on its next chunk. Revoking an already-expired token is a harmless no-op.

Parameters

Parameter	Type
`request`	`RevokeClientTokenRequest`

Returns

Promise<void>

Throws

when the credentials or token are invalid (401).

StaticTokenProvider

A TokenProvider that always returns the same pre-issued token.

Implements

TokenProvider

Constructors

Constructor

new StaticTokenProvider(token): StaticTokenProvider;

Parameters

Parameter	Type
`token`	`string`

Returns

StaticTokenProvider

Methods

getToken()

getToken(): Promise<string>;

Return a currently-valid bearer token, minting/refreshing as needed.

Returns

Promise<string>

Implementation of

TokenProvider.getToken

TokenManager

Mints access tokens from API-key credentials via AuthClient and caches them until shortly before they expire, so callers can request a valid bearer token cheaply on every call. Concurrent requests while a token is being minted share a single in-flight request.

Implements

TokenProvider

Constructors

Constructor

new TokenManager(options): TokenManager;

Parameters

Parameter	Type
`options`	`TokenManagerOptions`

Returns

TokenManager

Methods

getToken()

getToken(): Promise<string>;

Return a valid bearer token, minting or refreshing transparently.

Returns

Promise<string>

Implementation of

TokenProvider.getToken

invalidate()

invalidate(): void;

Drop any cached token so the next getToken mints a fresh one.

Returns

void

InterhumanClient

The main entry point. Construct it once with credentials (or a token) and use upload, stream, and realtime to call the API; token lifecycle is handled for you.

Example

const client = new InterhumanClient({
  credentials: { keyId: "...", keySecret: "..." },
});
const report = await client.upload.analyze({ file: bytes, filename: "clip.mp4" });

Constructors

Constructor

new InterhumanClient(options): InterhumanClient;

Parameters

Parameter	Type
`options`	`InterhumanClientOptions`

Returns

InterhumanClient

Properties

Property	Modifier	Type	Description
`auth`	`readonly`	`AuthClient`	Low-level auth client for minting tokens directly via `/v1/auth`.
`upload`	`readonly`	`UploadClient`	Upload-mode analysis client (`POST /v1/upload/analyze`).

Methods

stream()

stream(options?): StreamClient;

Create a new StreamClient for a live analysis session. Each call returns a fresh client (one per WebSocket session); register handlers and then call connect(). By default the session opens WS /v1/stream/analyze (Inter-1). Pass { apiVersion: "v2" } to open WS /v2/stream/analyze and have the same session analyzed by the Inter-2 model — the protocol, handlers and types are identical.

Parameters

Parameter	Type
`options`	`StreamOptions`

Returns

StreamClient

realtime()

realtime(): RealtimeClient;

Create a new RealtimeClient for a realtime multi-track analysis session (WS /v0/realtime/analyze). Each call returns a fresh client (one per WebSocket session); register handlers and then call connect(). The endpoint requires tokens holding the realtime scope, which is not in the client’s default scopes, so pass it explicitly (e.g. scopes: [Scope.Upload, Scope.Stream, Scope.Realtime]). Request every scope the client needs: the minted token grants exactly what was asked for, so narrowing to [Scope.Realtime] alone closes off upload() and stream().

Returns

RealtimeClient

getToken()

getToken(): Promise<string>;

Return a currently-valid bearer token (minting/refreshing as needed).

Returns

Promise<string>

InterhumanError

Base class for every error the SDK throws.

Extends

Error

Constructors

Constructor

new InterhumanError(message): InterhumanError;

Parameters

Parameter	Type
`message`	`string`

Returns

InterhumanError

Overrides

Error.constructor

InterhumanApiError

An error response from an HTTP endpoint (/v1/auth, /v1/upload/analyze). errorId, correlationId, and link are surfaced from the canonical ApiErrorBody when the server returned one.

Extends

InterhumanError

Constructors

Constructor

new InterhumanApiError(args): InterhumanApiError;

Parameters

Parameter	Type
`args`	{ `status`: `number`; `body?`: `ApiErrorBody`; `message?`: `string`; }
`args.status`	`number`
`args.body?`	`ApiErrorBody`
`args.message?`	`string`

Returns

InterhumanApiError

Overrides

InterhumanError.constructor

Properties

Property	Modifier	Type	Description
`status`	`readonly`	`number`	HTTP status code.
`errorId?`	`readonly`	`string`	Machine-readable error code, when present.
`correlationId?`	`readonly`	`string`	Correlation id for support, when present.
`link?`	`readonly`	`string`	Documentation link for the error, when present.
`body?`	`readonly`	`ApiErrorBody`	The raw parsed error body, when the server returned JSON.

UploadJobTimeoutError

An upload job did not reach a terminal state within the wait the caller allowed (UploadClient.waitForJob with timeoutMs). The job itself keeps running; job is the last envelope read, so the caller can keep polling with its job_id.

Extends

InterhumanError

Constructors

Constructor

new UploadJobTimeoutError(message, job): UploadJobTimeoutError;

Parameters

Parameter	Type
`message`	`string`
`job`	`unknown`

Returns

UploadJobTimeoutError

Overrides

InterhumanError.constructor

Properties

Property	Modifier	Type	Description
`job`	`readonly`	`unknown`	The last job envelope read before the wait gave up.

InterhumanConfigError

A configuration/usage error caught before any network call. Stream error envelopes are not thrown — they are delivered as typed StreamErrorEvents through the stream client’s on("error", …) surface, and transport-level socket failures arrive on "socketError" as a base InterhumanError.

Extends

InterhumanError

Constructors

Constructor

new InterhumanConfigError(message): InterhumanConfigError;

Parameters

Parameter	Type
`message`	`string`

Returns

InterhumanConfigError

Inherited from

InterhumanError.constructor

RealtimeClient

A typed client over the realtime WebSocket protocol. The realtime endpoint shares the stream endpoint’s session mechanics (subprotocol bearer auth, binary video frames, JSON envelopes down, and the requestClose() graceful shutdown handshake) and adds multi-track analysis, a caller-supplied transcript (sendTranscript), and a periodic recommendation step (realtime_recommendation.generated) enabled through the session config’s realtime_recommendation_instructions. The endpoint currently ships under the v0 path (/v0/realtime/analyze), which may change when it graduates to v1, and requires credentials holding the interhumanai.realtime scope.

Example

const realtime = client.realtime();
realtime.on("signal.detected", (e) => console.log(e.data.signal_type));
realtime.on("realtime_recommendation.generated", (e) => console.log(e.data.text));
await realtime.connect();
await realtime.waitForSessionReady();
realtime.updateConfig({
  realtime_recommendation_instructions: "Goal: help me close a sales call.",
  realtime_recommendation_frequency: "medium",
});
realtime.sendVideo(chunk);
realtime.requestClose(); // graceful: drains (incl. recommendation), then session.ended + close

Extends

SessionSocketClient<RealtimeEventMap, RealtimeSessionConfig>

Constructors

Constructor

new RealtimeClient(options): RealtimeClient;

Parameters

Parameter	Type
`options`	`SessionSocketClientOptions`

Returns

RealtimeClient

Overrides

SessionSocketClient.constructor

Properties

Property	Modifier	Type	Default value	Description	Overrides	Inherited from
`endpointPath`	`readonly`	`"/v0/realtime/analyze"`	`"/v0/realtime/analyze"`	Endpoint path appended to the WebSocket base URL (e.g. `/v1/stream/analyze`).	`SessionSocketClient`.`endpointPath`	-
`clientName`	`readonly`	`"RealtimeClient"`	`"RealtimeClient"`	How configuration error messages name this client. Usually the concrete class name; a client serving several endpoints may qualify it.	`SessionSocketClient`.`clientName`	-
`displayName`	`readonly`	`"Realtime"`	`"Realtime"`	Human-readable surface name used in error messages (e.g. `"Stream"`).	`SessionSocketClient`.`displayName`	-
`scopeHint`	`readonly`	`"The credentials may be invalid or lack the realtime scope."`	`"The credentials may be invalid or lack the realtime scope."`	Appended to the connect-rejection message to hint at the likely scope issue.	`SessionSocketClient`.`scopeHint`	-
`frameNoun?`	`readonly`	`string`	`undefined`	How frame-decoding messages name the surface mid-sentence (e.g. `"stream"` in “Received a malformed stream frame.”). Defaults to the lowercased displayName; set it where that would mangle a proper noun.	-	`SessionSocketClient`.`frameNoun`
`handshakeQuery?`	`readonly`	`Readonly`<`Record`<`string`, `string`>>	`undefined`	Query parameters carried on the upgrade request, beside the SDK identity pair. Unset for a client whose endpoint takes none; the stream client sets it to name the model a `/v2` session opens on.	-	`SessionSocketClient`.`handshakeQuery`

Accessors

isOpen

Get Signature

get isOpen(): boolean;

Whether the underlying socket is open.

Returns

boolean

Inherited from

SessionSocketClient.isOpen

Methods

sendTranscript()

sendTranscript(transcript): void;

Send (or replace) the caller-supplied transcript as a transcript.updated text frame. The most recent transcript replaces any prior one for the session and is rendered into the recommendation prompt. A transcript that does not match the segment shape is rejected by the server with a non-fatal error envelope; the session stays open.

Parameters

Parameter	Type
`transcript`	`TranscriptSegment`[]

Returns

void

on()

on<K>(event, handler): () => void;

Subscribe to an event. Returns an unsubscribe function. Use a specific envelope type (e.g. "signal.detected"), "message" for every envelope, or a lifecycle event ("open", "close", "socketError").

Type Parameters

Type Parameter
`K` extends keyof `RealtimeEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`RealtimeEventMap`[`K`]>

Returns

() => void

Inherited from

SessionSocketClient.on

off()

off<K>(event, handler): void;

Unsubscribe a previously registered handler.

Type Parameters

Type Parameter
`K` extends keyof `RealtimeEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`RealtimeEventMap`[`K`]>

Returns

void

Inherited from

SessionSocketClient.off

once()

once<K>(event, handler): () => void;

Subscribe to the next occurrence of an event, then auto-unsubscribe.

Type Parameters

Type Parameter
`K` extends keyof `RealtimeEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`RealtimeEventMap`[`K`]>

Returns

() => void

Inherited from

SessionSocketClient.once

connect()

connect(): Promise<void>;

Open the connection and resolve once it is established. Rejects if the connection closes or errors before opening (e.g. invalid credentials).

Returns

Promise<void>

Inherited from

SessionSocketClient.connect

waitForSessionReady()

waitForSessionReady(): Promise<RealtimeSessionReadyEvent>;

Wait for the session.ready envelope. Resolves immediately if it has already arrived (the envelope is retained), otherwise on the next one. Safe to call after connect without racing the frame. Rejects with an InterhumanError if the connection closes before the session becomes ready — including when the server accepts the handshake and then refuses the session, which it reports as an error envelope followed by a close (for example ih1003 when the endpoint’s backend is not configured). The message names the endpoint, the close code and reason, and that error envelope when there was one. A call made after such a close rejects immediately rather than waiting forever. Because the promise rejects rather than hanging, every caller must handle the rejection: a caller that races it against a timeout, or requests it before connect and awaits connect() first, should attach a .catch() so a refused session is not an unhandled rejection.

Returns

Promise<RealtimeSessionReadyEvent>

Inherited from

SessionSocketClient.waitForSessionReady

sendVideo()

sendVideo(chunk): void;

Send a chunk of WebM or fragmented-MP4 video as a binary frame.

Parameters

Parameter	Type
`chunk`	`VideoChunk`

Returns

void

Inherited from

SessionSocketClient.sendVideo

sendBinary()

protected sendBinary(payload): void;

Send an arbitrary binary frame.

Parameters

Parameter	Type
`payload`	\| `Blob` \| `ArrayBuffer` \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`> \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`>

Returns

void

Inherited from

SessionSocketClient.sendBinary

updateConfig()

updateConfig(config): void;

Send (or replace) the session config. Each frame fully replaces the active config; the server acknowledges with a session.updated envelope.

Parameters

Parameter	Type
`config`	`RealtimeSessionConfig`

Returns

void

Inherited from

SessionSocketClient.updateConfig

requestClose()

requestClose(): void;

Request a graceful end-of-analysis shutdown (session.close). The server acknowledges with a session.closing envelope whose data.max_drain_seconds is the longest you should wait for the final session.ended envelope. After the acknowledgment the server rejects new video, finishes analyzing the video it already accepted (emitting the normal envelopes in order, plus signal.ended for still-active signals), sends session.ended, and closes the socket with code 1000 — so listen for "session.ended" (or "close") rather than calling close.

Returns

void

Inherited from

SessionSocketClient.requestClose

close()

close(code?, reason?): void;

Close the connection immediately, without the graceful drain handshake.

Parameters

Parameter	Type
`code?`	`number`
`reason?`	`string`

Returns

void

Inherited from

SessionSocketClient.close

sendJson()

protected sendJson(payload): void;

Send an arbitrary JSON payload as a text frame.

Parameters

Parameter	Type
`payload`	`unknown`

Returns

void

Inherited from

SessionSocketClient.sendJson

`abstract` SessionSocketClient

Base class implementing the shared live-session WebSocket protocol. Authentication uses the Sec-WebSocket-Protocol: access_token, <token> subprotocol pair, which is portable across browsers and Node (the standard WebSocket constructor cannot set an Authorization header). Inbound frames are decoded into the protocol’s envelope union and dispatched through the typed on surface; outbound video is sent as binary frames and JSON payloads (session config, and for realtime the transcript) as text frames.

Extended by

StreamClient
RealtimeClient

Type Parameters

Type Parameter	Description
`TEventMap` extends `SessionSocketEventMapBase`<`unknown`, `unknown`>	Event payloads keyed by event name (envelope `type`s plus the lifecycle events of SessionSocketEventMapBase).
`TConfig` extends `object`	The client→server session-config shape.

Constructors

Constructor

new SessionSocketClient<TEventMap, TConfig>(options): SessionSocketClient<TEventMap, TConfig>;

Parameters

Parameter	Type
`options`	`SessionSocketClientOptions`

Returns

SessionSocketClient<TEventMap, TConfig>

Properties

Property	Modifier	Type	Description
`endpointPath`	`abstract`	`string`	Endpoint path appended to the WebSocket base URL (e.g. `/v1/stream/analyze`).
`clientName`	`abstract`	`string`	How configuration error messages name this client. Usually the concrete class name; a client serving several endpoints may qualify it.
`displayName`	`abstract`	`string`	Human-readable surface name used in error messages (e.g. `"Stream"`).
`frameNoun?`	`readonly`	`string`	How frame-decoding messages name the surface mid-sentence (e.g. `"stream"` in “Received a malformed stream frame.”). Defaults to the lowercased displayName; set it where that would mangle a proper noun.
`scopeHint`	`abstract`	`string`	Appended to the connect-rejection message to hint at the likely scope issue.
`handshakeQuery?`	`readonly`	`Readonly`<`Record`<`string`, `string`>>	Query parameters carried on the upgrade request, beside the SDK identity pair. Unset for a client whose endpoint takes none; the stream client sets it to name the model a `/v2` session opens on.

Accessors

isOpen

Get Signature

get isOpen(): boolean;

Whether the underlying socket is open.

Returns

boolean

Methods

on()

on<K>(event, handler): () => void;

Type Parameters

Type Parameter
`K` extends `string` \| `number` \| `symbol`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`TEventMap`[`K`]>

Returns

() => void

off()

off<K>(event, handler): void;

Unsubscribe a previously registered handler.

Type Parameters

Type Parameter
`K` extends `string` \| `number` \| `symbol`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`TEventMap`[`K`]>

Returns

void

once()

once<K>(event, handler): () => void;

Subscribe to the next occurrence of an event, then auto-unsubscribe.

Type Parameters

Type Parameter
`K` extends `string` \| `number` \| `symbol`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`TEventMap`[`K`]>

Returns

() => void

connect()

connect(): Promise<void>;

Open the connection and resolve once it is established. Rejects if the connection closes or errors before opening (e.g. invalid credentials).

Returns

Promise<void>

waitForSessionReady()

waitForSessionReady(): Promise<TEventMap["session.ready"]>;

Returns

Promise<TEventMap["session.ready"]>

sendVideo()

sendVideo(chunk): void;

Send a chunk of WebM or fragmented-MP4 video as a binary frame.

Parameters

Parameter	Type
`chunk`	`VideoChunk`

Returns

void

sendBinary()

protected sendBinary(payload): void;

Send an arbitrary binary frame.

Parameters

Parameter	Type
`payload`	\| `Blob` \| `ArrayBuffer` \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`> \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`>

Returns

void

updateConfig()

updateConfig(config): void;

Send (or replace) the session config. Each frame fully replaces the active config; the server acknowledges with a session.updated envelope.

Parameters

Parameter	Type
`config`	`TConfig`

Returns

void

requestClose()

requestClose(): void;

Returns

void

close()

close(code?, reason?): void;

Close the connection immediately, without the graceful drain handshake.

Parameters

Parameter	Type
`code?`	`number`
`reason?`	`string`

Returns

void

sendJson()

protected sendJson(payload): void;

Send an arbitrary JSON payload as a text frame.

Parameters

Parameter	Type
`payload`	`unknown`

Returns

void

StreamClient

A typed client over the stream WebSocket protocol. Authentication uses the Sec-WebSocket-Protocol: access_token, <token> subprotocol pair, which is portable across browsers and Node (the standard WebSocket constructor cannot set an Authorization header). Inbound frames are decoded into the StreamEvent union and dispatched through the typed on surface inherited from SessionSocketClient; outbound video is sent as binary frames and session config as a JSON text frame. The client serves both stream endpoints — /v1/stream/analyze (Inter-1, the default) and /v2/stream/analyze (Inter-2) — selected by StreamClientOptions.apiVersion. The protocol is the same on both, and each names the endpoint it actually opened in the errors it raises: a v2 session reports the Inter-2 stream, a v1 session the Stream. On /v2/stream/analyze the model is a permission of its own: the credential must carry interhumanai.stream.<model> for the model the session opens on, named by StreamClientOptions.model (default "inter-2"). session.ready lists, under supported_session_config_options.model, the models the deployment serves that the credential may select, and a later updateConfig({ model }) switching to one the credential lacks is answered with a non-fatal error envelope (ih2003) while the session continues on its current model. Add an audio_format on "inter-2-audio" to stream raw PCM frames with sendAudio instead of container media.

Examples

const stream = client.stream();               // Inter-1, /v1/stream/analyze
const streamV2 = client.stream({ apiVersion: "v2" }); // Inter-2, /v2/stream/analyze
stream.on("signal.detected", (e) => console.log(e.data.signal_type));
await stream.connect();
await stream.waitForSessionReady();
stream.updateConfig({ include: ["conversation_quality_overall"] });
stream.sendVideo(chunk);
stream.requestClose(); // graceful: drains, then session.ended + close

// Needs the interhumanai.stream.inter-2-audio scope.
const audio = client.stream({ apiVersion: "v2", model: "inter-2-audio" });
await audio.connect();
await audio.waitForSessionReady();
audio.updateConfig({
  audio_format: { encoding: "pcm_s16le", sample_rate: 16000, channels: 1 },
});
audio.sendAudio(pcmFrame); // signed 16-bit mono PCM, any frame size

Extends

SessionSocketClient<StreamEventMap, StreamSessionConfig>

Constructors

Constructor

new StreamClient(options): StreamClient;

Parameters

Parameter	Type
`options`	`StreamClientOptions`

Returns

StreamClient

Overrides

SessionSocketClient.constructor

Properties

Property	Modifier	Type	Description	Overrides
`endpointPath`	`readonly`	`string`	Endpoint path appended to the WebSocket base URL (e.g. `/v1/stream/analyze`).	`SessionSocketClient`.`endpointPath`
`clientName`	`readonly`	`string`	How configuration error messages name this client. Usually the concrete class name; a client serving several endpoints may qualify it.	`SessionSocketClient`.`clientName`
`displayName`	`readonly`	`string`	Human-readable surface name used in error messages (e.g. `"Stream"`).	`SessionSocketClient`.`displayName`
`frameNoun`	`readonly`	`string`	How frame-decoding messages name the surface mid-sentence (e.g. `"stream"` in “Received a malformed stream frame.”). Defaults to the lowercased displayName; set it where that would mangle a proper noun.	`SessionSocketClient`.`frameNoun`
`scopeHint`	`readonly`	`string`	Appended to the connect-rejection message to hint at the likely scope issue.	`SessionSocketClient`.`scopeHint`
`handshakeQuery?`	`readonly`	`Readonly`<`Record`<`string`, `string`>>	Query parameters carried on the upgrade request, beside the SDK identity pair. Unset for a client whose endpoint takes none; the stream client sets it to name the model a `/v2` session opens on.	`SessionSocketClient`.`handshakeQuery`
`apiVersion`	`readonly`	`StreamApiVersion`	The stream endpoint this client opens.	-
`model`	`readonly`	`StreamModel` \| `undefined`	The model a `"v2"` session opens on, as named at construction; `undefined` on `"v1"` and when the server default (`"inter-2"`) is left to apply.	-

Accessors

isOpen

Get Signature

get isOpen(): boolean;

Whether the underlying socket is open.

Returns

boolean

Inherited from

SessionSocketClient.isOpen

Methods

on()

on<K>(event, handler): () => void;

Type Parameters

Type Parameter
`K` extends keyof `StreamEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`StreamEventMap`[`K`]>

Returns

() => void

Inherited from

SessionSocketClient.on

off()

off<K>(event, handler): void;

Unsubscribe a previously registered handler.

Type Parameters

Type Parameter
`K` extends keyof `StreamEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`StreamEventMap`[`K`]>

Returns

void

Inherited from

SessionSocketClient.off

once()

once<K>(event, handler): () => void;

Subscribe to the next occurrence of an event, then auto-unsubscribe.

Type Parameters

Type Parameter
`K` extends keyof `StreamEventMap`

Parameters

Parameter	Type
`event`	`K`
`handler`	`Listener`<`StreamEventMap`[`K`]>

Returns

() => void

Inherited from

SessionSocketClient.once

connect()

connect(): Promise<void>;

Open the connection and resolve once it is established. Rejects if the connection closes or errors before opening (e.g. invalid credentials).

Returns

Promise<void>

Inherited from

SessionSocketClient.connect

waitForSessionReady()

waitForSessionReady(): Promise<SessionReadyEvent>;

Returns

Promise<SessionReadyEvent>

Inherited from

SessionSocketClient.waitForSessionReady

sendVideo()

sendVideo(chunk): void;

Send a chunk of WebM or fragmented-MP4 video as a binary frame.

Parameters

Parameter	Type
`chunk`	`VideoChunk`

Returns

void

Inherited from

SessionSocketClient.sendVideo

sendBinary()

protected sendBinary(payload): void;

Send an arbitrary binary frame.

Parameters

Parameter	Type
`payload`	\| `Blob` \| `ArrayBuffer` \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`> \| `Uint8Array`<`ArrayBufferLike`> \| `ArrayBufferView`<`ArrayBufferLike`>

Returns

void

Inherited from

SessionSocketClient.sendBinary

updateConfig()

updateConfig(config): void;

Send (or replace) the session config. Each frame fully replaces the active config; the server acknowledges with a session.updated envelope.

Parameters

Parameter	Type
`config`	`StreamSessionConfig`

Returns

void

Inherited from

SessionSocketClient.updateConfig

requestClose()

requestClose(): void;

Returns

void

Inherited from

SessionSocketClient.requestClose

close()

close(code?, reason?): void;

Close the connection immediately, without the graceful drain handshake.

Parameters

Parameter	Type
`code?`	`number`
`reason?`	`string`

Returns

void

Inherited from

SessionSocketClient.close

sendJson()

protected sendJson(payload): void;

Send an arbitrary JSON payload as a text frame.

Parameters

Parameter	Type
`payload`	`unknown`

Returns

void

Inherited from

SessionSocketClient.sendJson

sendAudio()

sendAudio(frame): void;

Send one frame of raw PCM audio as a binary frame. Only meaningful on apiVersion: "v2" after the session config declared an audio_format with model: "inter-2-audio"; the server then reads every binary frame as signed 16-bit mono PCM at the declared sample rate. Frames can be any size and cut at any byte boundary. Without that declaration a binary frame is container media, which sendVideo names.

Parameters

Parameter	Type
`frame`	`AudioFrame`

Returns

void

UploadClient

Analyzes complete files. analyze is the v1 route: the file is analyzed inside the request and the report comes back with the response. submit, getJob and waitForJob are the v2 job routes: a file is accepted as a job, analyzed by an Inter-2 model after the response, and read back by id until it is completed or failed. Every request is sent with the bearer token attached automatically from the configured TokenProvider.

Constructors

Constructor

new UploadClient(options): UploadClient;

Parameters

Parameter	Type
`options`	`UploadClientOptions`

Returns

UploadClient

Methods

analyze()

analyze(input): Promise<AnalysisResult>;

Analyze a video file in upload mode.

Parameters

Parameter	Type
`input`	`AnalyzeUploadInput`

Returns

Promise<AnalysisResult>

Throws

on a rejected request (auth, validation, too-large/too-short media, quota, unprocessable video, server error).

submit()

submit(input): Promise<UploadJob>;

Submit a file to POST /v2/upload/analyze as an asynchronous job. The API validates the file and answers with the job envelope before the analysis runs. With waitSeconds above zero the request is held open for up to that long, and the envelope comes back terminal (result or error set) when the job finished in time; otherwise it comes back queued or running, and waitForJob or getJob reads it later. Check status rather than assuming either.

Parameters

Parameter	Type
`input`	`SubmitUploadJobInput`

Returns

Promise<UploadJob>

Throws

on a rejected submit: ih4020 for a model the route does not serve yet, ih5008 for a file with no audio stream, ih4002 for an unsupported container, ih4007 for media under 3 s, ih1003 when no backend serves the model on the deployment, and the usual auth, size and quota errors.

getJob()

getJob(job, options?): Promise<UploadJob>;

Read a job’s current envelope from GET /v2/upload/jobs/{job_id}.

Parameters

Parameter	Type	Description
`job`	`string` \| `UploadJob`	The job’s id, or the envelope submit returned.
`options`	{ `signal?`: `AbortSignal`; }	-
`options.signal?`	`AbortSignal`	-

Returns

Promise<UploadJob>

Throws

ih4021 (404) when the id is unknown to this account or the job has expired, among the usual errors.

waitForJob()

waitForJob(job, options?): Promise<UploadJob>;

Poll a job until it is completed or failed, and resolve with it. A terminal envelope passed in resolves at once without a request. A failed job resolves, not rejects: read its error.

Parameters

Parameter	Type
`job`	`string` \| `UploadJob`
`options`	`WaitForJobOptions`

Returns

Promise<UploadJob>

Throws

when timeoutMs elapses first. The job keeps running; the error carries the last envelope read.

Interfaces

AuthClientOptions

Options for constructing an AuthClient.

Properties

Property	Type	Description
`baseUrl?`	`string`	Explicit base URL (e.g. `http://localhost:8080`). Overrides `environment`.
`environment?`	`Environment`	Named environment to target. Defaults to `"production"`.
`fetch?`	{ (`input`, `init?`): `Promise`<`Response`>; (`input`, `init?`): `Promise`<`Response`>; }	Custom `fetch` implementation. Defaults to the global `fetch`.

TokenProvider

Anything that can supply a bearer token on demand.

Methods

getToken()

getToken(): Promise<string>;

Return a currently-valid bearer token, minting/refreshing as needed.

Returns

Promise<string>

TokenManagerOptions

Options for a TokenManager.

Properties

Property	Type	Description
`authClient`	`AuthClient`	-
`credentials`	`ApiKeyCredentials`	-
`scopes`	`ScopeValue`[]	Scopes to request when minting tokens.
`refreshSkewSeconds?`	`number`	Refresh a cached token this many seconds before its stated expiry, to absorb clock skew and request latency. Defaults to 30.
`now?`	() => `number`	Injectable clock (milliseconds). Defaults to `Date.now`. Test seam.

ApiKeyCredentials

Credentials issued in the Interhuman customer platform.

Extended by

TokenRequest

Properties

Property	Type	Description
`keyId`	`string`	API key ID from the dashboard.
`keySecret`	`string`	API key secret from the dashboard.

ApiKeyCredential

A full API key string used to mint and revoke client tokens.

Extended by

ClientTokenRequest
RevokeClientTokenRequest

Properties

Property	Type	Description
`apiKey`	`string`	Your full API key from the dashboard (`ih_<env>_<publicId>_<entropy>`).

TokenRequest

Request body for POST /v1/auth.

Extends

ApiKeyCredentials

Properties

Property	Type	Description	Inherited from
`keyId`	`string`	API key ID from the dashboard.	`ApiKeyCredentials`.`keyId`
`keySecret`	`string`	API key secret from the dashboard.	`ApiKeyCredentials`.`keySecret`
`scopes`	`ScopeValue`[]	Requested scopes; at least one is required.	-

TokenResponse

Response body from POST /v1/auth.

Properties

Property	Type	Description
`access_token`	`string`	The minted bearer access token (a JWT).
`token_type`	`string`	Token type; always `"Bearer"`.
`expires_in`	`number`	Seconds until the token expires (typically 900).
`scope`	`string`	Space-separated list of granted scopes.

ClientTokenRequest

Request for minting an ephemeral, capped client token (POST /v1/client_tokens). Call this server-side with your API key; the returned token is safe to hand to a browser to call the upload, stream, or realtime endpoints directly. Every cap is optional. The token grants only the requested scopes (default interhumanai.stream); the caps are enforced on streaming sessions (stream / realtime) and the video budget additionally spans upload.

Extends

ApiKeyCredential

Properties

Property	Type	Description	Inherited from
`apiKey`	`string`	Your full API key from the dashboard (`ih_<env>_<publicId>_<entropy>`).	`ApiKeyCredential`.`apiKey`
`scopes?`	`ScopeValue`[]	Scopes to grant the token. Defaults to `["interhumanai.stream"]`.	-
`expiresIn?`	`number`	Requested lifetime in seconds. Clamped to 60–3600; defaults to 300.	-
`maxDurationSeconds?`	`number`	Max wall-clock duration, in seconds, of a session opened with this token.	-
`maxBytes?`	`number`	Max cumulative video bytes a session opened with this token may send.	-
`maxConcurrent?`	`number`	Max number of concurrent sessions opened with this token. Defaults to 1 (one session at a time) when omitted server-side.	-
`maxVideoSeconds?`	`number`	Max total seconds of video the token may process across upload/stream/realtime.	-
`allowedOrigins?`	`string`[]	Allow-list of browser `Origin` values permitted to use this token.	-

ClientTokenResponse

Response body from POST /v1/client_tokens.

Properties

Property	Type	Description
`access_token`	`string`	The minted client token (a JWT).
`token_type`	`string`	Token type; always `"Bearer"`.
`expires_in`	`number`	Seconds until the token expires.
`scope`	`string`	Space-separated list of granted scopes.
`max_duration_seconds?`	`number` \| `null`	Per-token max session duration that will be enforced, if any.
`max_bytes?`	`number` \| `null`	Per-token max cumulative session bytes that will be enforced, if any.
`max_concurrent?`	`number` \| `null`	Per-token max concurrent sessions that will be enforced, if any.
`max_video_seconds?`	`number` \| `null`	Per-token total video-seconds budget that will be enforced, if any.
`allowed_origins?`	`string`[] \| `null`	Per-token allow-list of browser origins that will be enforced, if any.

RevokeClientTokenRequest

Request for revoking a client token (POST /v1/client_tokens/revoke).

Extends

ApiKeyCredential

Properties

Property	Type	Description	Inherited from
`apiKey`	`string`	Your full API key from the dashboard (`ih_<env>_<publicId>_<entropy>`).	`ApiKeyCredential`.`apiKey`
`token`	`string`	The client token to revoke (as returned in `access_token`).	-

InterhumanClientOptions

Options for constructing an InterhumanClient.

Properties

Property	Type	Description
`credentials?`	`ApiKeyCredentials`	API-key credentials. The client exchanges these at `/v1/auth` and refreshes the resulting token automatically. Provide either `credentials` or `accessToken`.
`accessToken?`	`string`	A pre-issued bearer access token, used as-is. Provide either `accessToken` or `credentials`.
`scopes?`	`ScopeValue`[]	Scopes requested when minting tokens from `credentials`. Defaults to upload + stream. Ignored when `accessToken` is supplied.
`environment?`	`Environment`	Named environment to target. Defaults to `"production"`.
`baseUrl?`	`string`	Explicit base URL (e.g. `http://localhost:8080`). Overrides `environment`.
`fetch?`	{ (`input`, `init?`): `Promise`<`Response`>; (`input`, `init?`): `Promise`<`Response`>; }	Custom `fetch` implementation. Defaults to the global `fetch`.
`webSocket?`	`WebSocketFactory`	Custom WebSocket factory for the stream and realtime clients. Defaults to the global `WebSocket` (browsers, Node 22+).
`refreshSkewSeconds?`	`number`	Seconds before stated expiry to refresh a minted token. Defaults to 30.

StreamOptions

Per-session options for InterhumanClient.stream.

Properties

Property	Type	Description
`apiVersion?`	`StreamApiVersion`	Which stream endpoint to open: `"v1"` (Inter-1, the default) or `"v2"` (Inter-2). Both endpoints speak the same protocol; only the model behind the analysis differs, and with it the scope: `interhumanai.stream` for v1, `interhumanai.stream.<model>` for v2.
`model?`	`StreamModel`	The Inter-2 model a `"v2"` session opens on (default `"inter-2"`); the credential must carry `interhumanai.stream.<model>` for it. See StreamClientOptions.model.

ApiErrorBody

The canonical error body returned by every HTTP error path.

Properties

Property	Type	Description
`error_id`	`string`	Machine-readable error code.
`correlation_id?`	`string` \| `null`	Connection/request correlation id; quote it when contacting support.
`link?`	`string` \| `null`	URL with more information about this error, when available.
`message?`	`string` \| `null`	Human-readable explanation, when available.

TranscriptSegment

One diarized utterance of a realtime transcript, as sent with sendTranscript.

Properties

Property	Type	Description
`start`	`number`	Start time of the utterance, in seconds.
`end`	`number`	End time of the utterance, in seconds.
`text`	`string`	The spoken text of this segment.
`speaker`	`number`	Zero-based index identifying the speaker of this segment.

RealtimeSessionConfig

Client-to-server realtime session config (sent as a JSON text frame). The last config sent fully replaces the active one; the server acknowledges with a session.updated envelope. The stream include flags are not accepted — the realtime endpoint never emits the conversation-quality or engagement updates those flags control.

Properties

Property	Type	Description
`analysis_groups?`	`AnalysisGroup`[]	Analysis surfaces to run for this session, any subset of what the server advertises on `session.ready` under `supported_session_config_options.analysis_groups`. Must be non-empty when present: an empty array — like selecting a group the server has not enabled — is rejected with an `error` envelope. When omitted, the server default is `audio` + `visual`. The `visual` group reads the picture, so select `["audio"]` alone to analyze a stream whose recording carries no video track. Leaving `visual` selected and sending such a stream is a configuration error rather than a partial success: the server sends one `ih5001` `error` envelope and stops analyzing the session, the audio tracks included. Sending a config that narrows to `["audio"]` resumes analysis from the next window.
`realtime_recommendation_frequency?`	`RealtimeRecommendationFrequency`	How often recommendation runs once enabled. Has no effect until a non-empty `realtime_recommendation_instructions` enables recommendations.
`realtime_recommendation_instructions?`	`string` \| `null`	Configuration block for the periodic recommendation step, and the switch that enables it: the recommendation step runs only when this is a non-empty string (there is no default prompt). It sets the goal, domain/context, and the focus, tone, or output style of the guidance. It is treated as untrusted configuration data — it shapes the guidance but cannot override the recommendation task or output contract set by the active system prompt.

TranscriptUpdatedFrame

The inbound transcript.updated text frame carrying a caller-supplied transcript. Sent by RealtimeClient.sendTranscript; the most recent transcript replaces any prior one and is rendered into the recommendation prompt.

Properties

Property	Type	Description
`type`	`"transcript.updated"`	-
`transcript`	`TranscriptSegment`[]	The full transcript as `{start, end, text, speaker}` segments, in spoken order.

RealtimeSessionReadyConfigOptions

Supported realtime session-config options advertised on session.ready. Unlike the stream options, only the realtime controls appear here — never the stream include / goal_dimensions options.

Properties

Property	Type	Description
`realtime_recommendation_instructions`	`"string"`	Always the literal `"string"`: `realtime_recommendation_instructions` takes free-form text, so its accepted JSON type is advertised instead of a value list.
`realtime_recommendation_frequency`	`RealtimeRecommendationFrequency`[]	Accepted `realtime_recommendation_frequency` values.
`analysis_groups`	`AnalysisGroup`[]	Analysis groups this deployment allows — the valid values for the config’s `analysis_groups` field, so you can pick a subset without round-tripping a rejection.

RealtimeSessionReadyData

session.ready payload — the same connection-level limits as the stream session.ready, with the realtime supported-config options.

Properties

Property	Type	Description
`session_idle_timeout_seconds`	`number`	Idle seconds tolerated before the server closes the session (`0` = disabled).
`session_max_duration_seconds`	`number`	Maximum total session duration in seconds (`0` = disabled).
`max_segment_duration_seconds`	`number` \| `null`	Max probed duration of one inbound chunk, or `null` when only the size cap binds.
`min_segment_size_bytes`	`number`	Minimum size in bytes of one inbound chunk.
`max_segment_size_bytes`	`number`	Maximum size in bytes of one inbound chunk.
`supported_session_config_options`	`RealtimeSessionReadyConfigOptions`	-

RealtimeSessionReadyEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`type`	`"session.ready"`	-	-
`data`	`RealtimeSessionReadyData`	-	-
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`

RealtimeSessionUpdatedData

session.updated payload — the consolidated realtime config after a config frame applied. Reports only the realtime options, never the stream-only include / goal_dimensions fields.

Properties

Property	Type	Description
`realtime_recommendation_instructions`	`string` \| `null`	The applied `realtime_recommendation_instructions`, echoed back verbatim; `null` when unset.
`realtime_recommendation_frequency`	\| `RealtimeRecommendationFrequency` \| `null`	The active recommendation frequency; `null` when unset.
`analysis_groups`	`AnalysisGroup`[]	The effective analysis-group selection in force (the server default when you have not chosen one).

RealtimeSessionUpdatedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`type`	`"session.updated"`	-	-
`data`	`RealtimeSessionUpdatedData`	-	-
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`

RealtimeRecommendationGeneratedData

realtime_recommendation.generated payload — one recommendation run’s guidance text.

Properties

Property	Type	Description
`text`	`string`	The guidance text returned by the recommendation model — one short recommendation for the responding speaker, or the literal `NO_GUIDANCE` when the signals warrant no adjustment.
`start`	`number`	Start of the analyzed interval this recommendation covers (seconds, session time).
`end`	`number`	End of the analyzed interval this recommendation covers (seconds, session time).

RealtimeRecommendationGeneratedEvent

realtime_recommendation.generated — the periodic recommendation result. Emitted only when realtime_recommendation_instructions has enabled recommendation, paced by realtime_recommendation_frequency.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`type`	`"realtime_recommendation.generated"`	-	-
`data`	`RealtimeRecommendationGeneratedData`	-	-
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`

RealtimeSignalDetectedData

Properties

Property	Type	Description
`signal_type`	`SignalType`	-
`start`	`number`	Seconds, absolute session-cumulative time.
`probability`	`Probability` \| `null`	-
`modality`	`string`[]	Analysis modalities that detected this signal, naming the source of the evidence: one or both of `audio` and `visual`. When both detect the same signal, both are included.

RealtimeSignalDetectedEvent

A social signal was detected. The realtime payload carries no rationale.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`type`	`"signal.detected"`	-	-
`data`	`RealtimeSignalDetectedData`	-	-
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`

RealtimeSignalUpdatedData

Properties

Property	Type	Description
`signal_type`	`SignalType`	-
`start`	`number`	Seconds, absolute session-cumulative time.
`probability`	`Probability` \| `null`	-
`modality`	`string`[]	Analyses whose evidence produced this signal. A change to this set is one of the things that emits `signal.updated`.

RealtimeSignalUpdatedEvent

An active signal’s details changed. The realtime payload carries no rationale.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`type`	`"signal.updated"`	-	-
`data`	`RealtimeSignalUpdatedData`	-	-
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`

RealtimeEventMap

Event payloads keyed by event name, for the typed on/off surface of the realtime client. The envelope types map to their envelope objects; the remaining keys are client-level lifecycle events.

Properties

Property	Type	Description
`session.ready`	`RealtimeSessionReadyEvent`	-
`session.updated`	`RealtimeSessionUpdatedEvent`	-
`session.closing`	`SessionClosingEvent`	-
`session.ended`	`SessionEndedEvent`	-
`signal.detected`	`RealtimeSignalDetectedEvent`	-
`signal.updated`	`RealtimeSignalUpdatedEvent`	-
`signal.ended`	`SignalEndedEvent`	-
`realtime_recommendation.generated`	`RealtimeRecommendationGeneratedEvent`	-
`coverage.dropped`	`CoverageDroppedEvent`	-
`coverage.degraded`	`CoverageDegradedEvent`	-
`error`	`StreamErrorEvent`	-
`message`	`RealtimeEvent`	Fires for every server→client envelope, regardless of type.
`open`	`void`	Fires once when the WebSocket connection opens.
`close`	`StreamCloseInfo`	Fires once when the connection closes.
`socketError`	`Error`	Transport-level socket error (distinct from the `error` envelope).

SessionSocketClientOptions

Options shared by every WebSocket session client.

Extended by

StreamClientOptions

Properties

Property	Type	Description
`tokenProvider`	`TokenProvider`	Supplies the bearer token used to authenticate the connection.
`baseUrl?`	`string`	Explicit base URL (e.g. `http://localhost:8080`). Overrides `environment`.
`environment?`	`Environment`	Named environment to target. Defaults to `"production"`.
`webSocket?`	`WebSocketFactory`	Custom WebSocket factory. Defaults to the global `WebSocket` (browsers, Node 22+). Supply one backed by the `ws` package on older Node.

SessionSocketEventMapBase

The lifecycle and wildcard events every session event map carries, alongside its protocol-specific envelope type keys. TEvent is the protocol’s envelope union and TReady its session.ready envelope.

Type Parameters

Type Parameter
`TEvent`
`TReady`

Properties

Property	Type	Description
`session.ready`	`TReady`	The `session.ready` envelope — first frame after the handshake.
`message`	`TEvent`	Fires for every server→client envelope, regardless of type.
`open`	`void`	Fires once when the WebSocket connection opens.
`close`	`StreamCloseInfo`	Fires once when the connection closes.
`socketError`	`Error`	Transport-level socket error (distinct from the `error` envelope).

StreamClientOptions

Options for constructing a StreamClient.

Extends

SessionSocketClientOptions

Properties

Property	Type	Description	Inherited from
`tokenProvider`	`TokenProvider`	Supplies the bearer token used to authenticate the connection.	`SessionSocketClientOptions`.`tokenProvider`
`baseUrl?`	`string`	Explicit base URL (e.g. `http://localhost:8080`). Overrides `environment`.	`SessionSocketClientOptions`.`baseUrl`
`environment?`	`Environment`	Named environment to target. Defaults to `"production"`.	`SessionSocketClientOptions`.`environment`
`webSocket?`	`WebSocketFactory`	Custom WebSocket factory. Defaults to the global `WebSocket` (browsers, Node 22+). Supply one backed by the `ws` package on older Node.	`SessionSocketClientOptions`.`webSocket`
`apiVersion?`	`StreamApiVersion`	Which stream endpoint to open. Defaults to `"v1"` (Inter-1). Pass `"v2"` to analyze the same session with the Inter-2 model at `/v2/stream/analyze`; everything else about the session is identical.	-
`model?`	`StreamModel`	The Inter-2 model a `"v2"` session opens on, sent as the handshake’s `model` query parameter. Defaults to `"inter-2"` when omitted. The credential must carry the scope for that model — `interhumanai.stream.inter-2` or `interhumanai.stream.inter-2-audio` — or the server refuses the session with `ih2003`. `"inter-2-deep"` is an upload model and is rejected here at construction; on `"v1"` the option is rejected too, since that endpoint offers no model selection.	-

StreamEnvelopeBase

Fields shared by every server→client envelope.

Extended by

ConversationQualityUpdatedEvent
CoverageDegradedEvent
CoverageDroppedEvent
EngagementUpdatedEvent
FeedbackGeneratedEvent
SessionClosingEvent
SessionEndedEvent
SessionReadyEvent
SessionUpdatedEvent
SignalDetectedEvent
SignalEndedEvent
SignalUpdatedEvent
StreamErrorEvent
RealtimeSessionReadyEvent
RealtimeSessionUpdatedEvent
RealtimeSignalDetectedEvent
RealtimeSignalUpdatedEvent
RealtimeRecommendationGeneratedEvent

Properties

Property	Type	Description
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.

RawAudioFormat

Declares that a session’s binary frames are raw audio rather than container media: signed 16-bit little-endian PCM at sample_rate, mono, with no header, cut at any byte boundary. Only "inter-2-audio" accepts it.

Properties

Property	Type
`encoding`	`"pcm_s16le"`
`sample_rate`	`PcmSampleRate`
`channels`	`1`

SessionReadyConfigOptions

Supported session-config options advertised on session.ready.

Properties

Property	Type	Description
`include`	`IncludeFlag`[]	-
`goal_dimensions`	`GoalDimension`[] \| `null`	`null` when the feedback feature is disabled.
`model`	`StreamModel`[] \| `null`	The models this deployment serves on `WS /v2/stream/analyze`; `null` on `WS /v1/stream/analyze`, which offers no model selection.

SessionReadyData

session.ready payload — session limits and supported config.

Properties

Property	Type	Description
`session_idle_timeout_seconds`	`number`	Idle seconds tolerated before the server closes the session (`0` = disabled).
`session_max_duration_seconds`	`number`	Maximum total session duration in seconds (`0` = disabled).
`max_segment_duration_seconds`	`number` \| `null`	Max probed duration of one inbound chunk, or `null` when only the size cap binds.
`min_segment_size_bytes`	`number`	Minimum size in bytes of one inbound chunk.
`max_segment_size_bytes`	`number`	Maximum size in bytes of one inbound chunk.
`supported_session_config_options`	`SessionReadyConfigOptions`	-

SessionReadyEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"session.ready"`	-	-
`data`	`SessionReadyData`	-	-

SessionUpdatedData

session.updated payload — consolidated config after a config frame applied.

Properties

Property	Type	Description
`include`	`IncludeFlag`[]	-
`goal_dimensions`	`GoalDimension`[] \| `null`	-
`model`	`StreamModel` \| `null`	The model analyzing the session, on `WS /v2/stream/analyze`; `null` on v1.

SessionUpdatedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"session.updated"`	-	-
`data`	`SessionUpdatedData`	-	-

SessionClosingData

session.closing payload — the graceful-close drain contract.

Properties

Property	Type	Description
`max_drain_seconds`	`number`	Maximum seconds the server will spend draining accepted work before it closes the session; `session.ended` arrives no later than this. The session closes earlier when the accepted work finishes sooner.

SessionClosingEvent

Acknowledges a caller-initiated session.close request. From this point the server rejects new video, finishes analyzing the video it already accepted, emits final lifecycle envelopes (signal.ended for still-active signals), sends session.ended, and closes the socket.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"session.closing"`	-	-
`data`	`SessionClosingData`	-	-

SessionEndedData

session.ended payload — why the session ended.

Properties

Property	Type	Description
`reason`	`"client_shutdown"`	`client_shutdown` = the caller requested a graceful close.

SessionEndedEvent

The final message of a gracefully closed session: all analysis is closed and the server closes the WebSocket (code 1000) immediately after sending it. No further analysis messages follow.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"session.ended"`	-	-
`data`	`SessionEndedData`	-	-

SignalDetectedData

Properties

Property	Type	Description
`signal_type`	`SignalType`	-
`start`	`number`	Seconds, absolute session-cumulative time.
`probability`	`Probability` \| `null`	-
`rationale`	`string` \| `null`	-
`modality`	`string`[]	Analysis modalities that detected this signal, naming the source of the evidence. For stream sessions this is `["video"]`.

SignalDetectedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"signal.detected"`	-	-
`data`	`SignalDetectedData`	-	-

SignalUpdatedData

Properties

Property	Type	Description
`signal_type`	`SignalType`	-
`start`	`number`	-
`probability`	`Probability` \| `null`	-
`rationale`	`string` \| `null`	-
`modality`	`string`[]	Analyses whose evidence produced this signal. A change to this set is one of the things that emits `signal.updated`.

SignalUpdatedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"signal.updated"`	-	-
`data`	`SignalUpdatedData`	-	-

SignalEndedData

Properties

Property	Type	Description
`signal_type`	`SignalType`	-
`end`	`number`	Seconds, absolute session-cumulative time.

SignalEndedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"signal.ended"`	-	-
`data`	`SignalEndedData`	-	-

EngagementUpdatedData

Properties

Property	Type
`state`	`EngagementLevel`
`start`	`number`

EngagementUpdatedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"engagement.updated"`	-	-
`data`	`EngagementUpdatedData`	-	-

ConversationQualityUpdatedData

Properties

Property	Type	Description
`overall`	`ConversationQualityValues` \| `null`	Non-null when `conversation_quality_overall` was requested.
`timeline`	\| `ConversationQualityTimelineEntry`[] \| `null`	Non-null when `conversation_quality_timeline` was requested and the period had signals.

ConversationQualityUpdatedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"conversation_quality.updated"`	-	-
`data`	`ConversationQualityUpdatedData`	-	-

FeedbackGeneratedData

Properties

Property	Type	Description
`feedback`	`string`	Comma-separated relevant signal types observed since the previous emission.

FeedbackGeneratedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"feedback.generated"`	-	-
`data`	`FeedbackGeneratedData`	-	-

CoverageDroppedRange

A contiguous range of session time no analysis covered (not billed).

Properties

Property	Type
`start`	`number`
`end`	`number`

CoverageDroppedData

Payload of CoverageDroppedEvent: the time ranges no analysis covers. Emitted when the analysis pipeline sheds buffered video under backpressure, and when the incoming stream itself skipped ahead — a client stall whose media never arrived while its recorder clock kept running. Either way the listed ranges were never analyzed and are not billed.

Properties

Property	Type
`ranges`	`CoverageDroppedRange`[]

CoverageDroppedEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"coverage.dropped"`	-	-
`data`	`CoverageDroppedData`	-	-

CoverageDegradedRange

One time range whose visual coverage was partial, in session seconds.

Properties

Property	Type
`start`	`number`
`end`	`number`

CoverageDegradedData

Payload of CoverageDegradedEvent: the affected time ranges plus the reason. Unlike coverage.dropped, these windows were analyzed (audio plus whatever video decoded) and are billed normally — only the visual coverage was partial.

Properties

Property	Type	Description
`ranges`	`CoverageDegradedRange`[]	-
`reason`	`"video_gap"`	Why the visual coverage was partial. Currently always `"video_gap"`: the windows’ video packets covered materially less than the window span (e.g. a keyframe interval longer than the analysis window, as a static screen share produces).

CoverageDegradedEvent

Informational notice that one or more analysis windows were analyzed with partial visual coverage. The session stays open and the windows are billed normally; the ranges tell your application that visual signals over that stretch drew on partial video.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"coverage.degraded"`	-	-
`data`	`CoverageDegradedData`	-	-

StreamErrorData

Properties

Property	Type	Description
`code`	`string`	Machine-readable error code (e.g. `"ih6002"`).
`message`	`string`	-
`link`	`string` \| `null`	-
`segment`	`number` \| `null`	Inbound caller-chunk index when applicable, else `null`.

StreamErrorEvent

Fields shared by every server→client envelope.

Extends

StreamEnvelopeBase

Properties

Property	Type	Description	Inherited from
`timestamp`	`string`	ISO 8601 timestamp identifying when the event occurred.	`StreamEnvelopeBase`.`timestamp`
`correlation_id`	`string`	Connection correlation id; quote it when contacting support.	`StreamEnvelopeBase`.`correlation_id`
`type`	`"error"`	-	-
`data`	`StreamErrorData`	-	-

StreamSessionConfig

Client-to-server session config (sent as a JSON text frame).

Properties

Property	Type	Description
`include?`	`IncludeFlag`[]	Conversation-quality sections to include in result messages.
`goal_dimensions?`	`GoalDimension`[]	Goal dimensions for this session (drive periodic feedback).
`model?`	`StreamModel`	The model to analyze the session with, on `WS /v2/stream/analyze` only. Unlike `include`, this is session state: a frame that omits it keeps the model in force, and a different model is accepted only before the first media frame. Defaults to `"inter-2"`.
`audio_format?`	`RawAudioFormat`	Declares that the session’s binary frames are raw PCM audio (see RawAudioFormat). Requires `model: "inter-2-audio"` and is fixed once the first frame has been sent. Omit it to send WebM or fragmented-MP4 media.

SessionCloseRequest

Client-to-server graceful close request (sent as a JSON text frame). Tells the server the caller is done sending video; the server acknowledges with session.closing and ends the session with session.ended.

Properties

Property	Type
`type`	`"session.close"`

StreamCloseInfo

Information about a closed stream connection.

Properties

Property	Type
`code`	`number`
`reason`	`string`

StreamEventMap

Event payloads keyed by event name, for the typed on/off surface. The twelve envelope types map to their envelope objects; the remaining keys are client-level lifecycle events.

Properties

Property	Type	Description
`session.ready`	`SessionReadyEvent`	-
`session.updated`	`SessionUpdatedEvent`	-
`session.closing`	`SessionClosingEvent`	-
`session.ended`	`SessionEndedEvent`	-
`signal.detected`	`SignalDetectedEvent`	-
`signal.updated`	`SignalUpdatedEvent`	-
`signal.ended`	`SignalEndedEvent`	-
`engagement.updated`	`EngagementUpdatedEvent`	-
`conversation_quality.updated`	`ConversationQualityUpdatedEvent`	-
`feedback.generated`	`FeedbackGeneratedEvent`	-
`coverage.dropped`	`CoverageDroppedEvent`	-
`coverage.degraded`	`CoverageDegradedEvent`	-
`error`	`StreamErrorEvent`	-
`message`	`StreamEvent`	Fires for every server→client envelope, regardless of type.
`open`	`void`	Fires once when the WebSocket connection opens.
`close`	`StreamCloseInfo`	Fires once when the connection closes.
`socketError`	`Error`	Transport-level socket error (distinct from the `error` envelope).

WebSocketLike

The subset of the WHATWG WebSocket interface the SDK relies on. Both the browser WebSocket and Node 22+‘s global WebSocket satisfy this.

Properties

Property	Modifier	Type
`readyState`	`readonly`	`number`
`binaryType`	`public`	`string`

Methods

send()

send(data): void;

Parameters

Parameter	Type
`data`	`string` \| `Blob` \| `ArrayBufferLike` \| `ArrayBufferView`<`ArrayBufferLike`>

Returns

void

close()

close(code?, reason?): void;

Parameters

Parameter	Type
`code?`	`number`
`reason?`	`string`

Returns

void

addEventListener()

addEventListener(type, listener): void;

Parameters

Parameter	Type
`type`	`string`
`listener`	(`event`) => `void`

Returns

void

removeEventListener()

removeEventListener(type, listener): void;

Parameters

Parameter	Type
`type`	`string`
`listener`	(`event`) => `void`

Returns

void

Signal

A single social signal detected over a span of the analyzed media.

Properties

Property	Type	Description
`type`	`SignalType`	-
`start`	`number`	Start time in seconds.
`end`	`number`	End time in seconds.
`probability`	`Probability` \| `null`	-
`rationale`	`string` \| `null`	-
`modality`	`string`[]	Analysis modalities that detected this signal, naming the source of the evidence. When several tracks detect the same signal, every contributing modality is included.

EngagementStateEntry

A contiguous stretch of media labeled with a single engagement state.

Properties

Property	Type
`state`	`EngagementLevel`
`start`	`number`
`end`	`number`

ConversationQualityValues

The five conversation-quality dimension scores plus their mean (0–100).

Properties

Property	Type
`quality_index`	`number`
`clarity`	`number`
`authority`	`number`
`energy`	`number`
`rapport`	`number`
`learning`	`number`

ConversationQualityTimelineEntry

Conversation-quality scores for one window of the timeline.

Properties

Property	Type
`start`	`number`
`end`	`number`
`values`	`ConversationQualityValues`

ConversationQuality

Overall and time-varying conversation-quality scores.

Properties

Property	Type
`overall`	`ConversationQualityValues`
`timeline`	`ConversationQualityTimelineEntry`[]

InteractionFeedback

Structured interaction feedback returned by the coach model.

Properties

Property	Type
`type`	`FeedbackType`
`message?`	`string` \| `null`
`active_dimensions?`	`GoalDimension`[]
`primary_signal?`	`SignalType` \| `null`
`reason?`	`string` \| `null`

AnalysisResult

The full report returned by POST /v1/upload/analyze.

Properties

Property	Type
`signals`	`Signal`[]
`engagement_state`	`EngagementStateEntry`[]
`feedback?`	`InteractionFeedback` \| `null`
`conversation_quality?`	`ConversationQuality` \| `null`

AnalyzeUploadInput

Arguments for UploadClient.analyze.

Properties

Property	Type	Description
`file`	`VideoInput`	The video file to analyze (mp4/avi/mov/mkv/mpeg-ts/webm, ≥3s, ≤32MB).
`filename?`	`string`	Filename used when `file` is a `Blob` without a name. Defaults to `"video"`.
`include?`	`IncludeFlag`[]	Optional response sections to include. When omitted, conversation-quality scores are not returned.
`goalDimensions?`	`GoalDimension`[]	Optional goal dimensions for this interaction. Supplying at least one triggers interaction feedback. Ignored when `conversationContext` is set.
`conversationContext?`	`string`	Optional free-text scenario description. When set, it is the sole source of context for feedback generation and `goalDimensions` is ignored.
`signal?`	`AbortSignal`	Optional `AbortSignal` to cancel the request.

UploadJobWindow

The analysis of one fixed-length window of an upload job’s file. Each signal carries the window’s span as its start and end, and its modality names the evidence the model read (["audio"] for inter-2-audio).

Properties

Property	Type	Description
`index`	`number`	Zero-based position of the window in the file.
`start_seconds`	`number`	Where the window starts, in seconds from the start of the file.
`end_seconds`	`number`	Where the window ends, in seconds from the start of the file.
`engagement_status`	`EngagementLevel`	-
`signals`	`Signal`[]	-

UploadJobResult

The result of a completed upload job: one entry per analyzed window.

Properties

Property	Type	Description
`duration_seconds`	`number`	Length of the analyzed media, in seconds.
`window_seconds`	`number`	Length of each analysis window, in seconds.
`windows`	`UploadJobWindow`[]	Per-window analyses, in file order.

UploadJobError

Why an upload job failed, in the API’s standard error-body shape.

Properties

Property	Type	Description
`error_id`	`string`	Machine-readable error code.
`correlation_id?`	`string` \| `null`	-
`link?`	`string` \| `null`	-
`message?`	`string` \| `null`	-

UploadJob

The job envelope returned by POST /v2/upload/analyze and GET /v2/upload/jobs/{job_id}. result is set only when status is completed; error only when it is failed.

Properties

Property	Type	Description
`job_id`	`string`	Identifier of the job; use it with UploadClient.getJob.
`status`	`UploadJobStatus`	-
`model`	`UploadModel`	-
`created_at`	`string`	When the job was accepted, as an ISO 8601 UTC timestamp.
`expires_at`	`string`	When the job stops being readable, as an ISO 8601 UTC timestamp.
`status_url`	`string`	Path of the status resource, relative to the API base URL.
`result?`	`UploadJobResult` \| `null`	-
`error?`	`UploadJobError` \| `null`	-

SubmitUploadJobInput

Arguments for UploadClient.submit.

Properties

Property	Type	Description
`file`	`VideoInput`	The file to analyze. For `inter-2-audio`: wav, flac, mp3, m4a, ogg, or a webm or mp4 with an audio track — at least 3 seconds of media, at most 32 MB, and no longer than the deployment’s maximum duration (30 minutes by default; a longer file rejects with `ih4004`).
`filename?`	`string`	Filename used when `file` is a `Blob` without a name. Defaults to `"video"`.
`model`	`UploadModel`	The Inter-2 model to analyze with. `inter-2` and `inter-2-deep` are valid values the route does not serve yet and reject with `ih4020`.
`waitSeconds?`	`number`	How long, in seconds, the API may hold the request open waiting for the job to finish. `0` (the default) answers with the queued envelope at once; when the job finishes within the wait, the envelope comes back terminal. The deployment bounds it; a larger value rejects with `ih4005`.
`signal?`	`AbortSignal`	Optional `AbortSignal` to cancel the request.

WaitForJobOptions

Options for UploadClient.waitForJob.

Properties

Property	Type	Description
`timeoutMs?`	`number`	Give up after this many milliseconds and reject with UploadJobTimeoutError. Omit to wait until the job is terminal.
`pollIntervalMs?`	`number`	Milliseconds between status reads. Defaults to 2000.
`signal?`	`AbortSignal`	Optional `AbortSignal` that stops polling and rejects each read.

UploadClientOptions

Options for constructing an UploadClient.

Properties

Property	Type	Description
`tokenProvider`	`TokenProvider`	Supplies the bearer token attached to each request.
`baseUrl?`	`string`	Explicit base URL (e.g. `http://localhost:8080`). Overrides `environment`.
`environment?`	`Environment`	Named environment to target. Defaults to `"production"`.
`fetch?`	{ (`input`, `init?`): `Promise`<`Response`>; (`input`, `init?`): `Promise`<`Response`>; }	Custom `fetch` implementation. Defaults to the global `fetch`.

Type Aliases

Environment

type Environment = "production" | "staging";

A named Interhuman API environment.

RealtimeClientOptions

type RealtimeClientOptions = SessionSocketClientOptions;

Options for constructing a RealtimeClient.

AnalysisGroup

type AnalysisGroup = "visual" | "audio";

Top-level analysis surfaces a realtime session can run. Each group maps to one or more server-side analysis tracks: visual is the visual-signals track and audio the audio tracks.

RealtimeRecommendationFrequency

type RealtimeRecommendationFrequency = "high" | "medium" | "low";

How often the recommendation step runs, measured in analyzed video: high = every 10 seconds, medium = every 20 seconds, low = every 30 seconds. Only takes effect once realtime_recommendation_instructions has enabled recommendation.

RealtimeEvent

type RealtimeEvent = 
  | RealtimeSessionReadyEvent
  | RealtimeSessionUpdatedEvent
  | SessionClosingEvent
  | SessionEndedEvent
  | RealtimeSignalDetectedEvent
  | RealtimeSignalUpdatedEvent
  | SignalEndedEvent
  | TranscriptGeneratedEvent
  | RealtimeRecommendationGeneratedEvent
  | CoverageDroppedEvent
  | CoverageDegradedEvent
  | StreamErrorEvent;

Discriminated union of every realtime server→client envelope.

VideoChunk

type VideoChunk = Uint8Array | ArrayBuffer | ArrayBufferView | Blob;

A chunk of WebM or fragmented-MP4 video to stream to the server.

AudioFrame

type AudioFrame = Uint8Array | ArrayBuffer | ArrayBufferView | Blob;

A frame of raw PCM audio to stream to the server.

Listener

type Listener<T> = (payload) => void;

An event handler registered via on/once.

Type Parameters

Type Parameter
`T`

Parameters

Parameter	Type
`payload`	`T`

Returns

void

StreamApiVersion

type StreamApiVersion = "v1" | "v2";

Which stream endpoint a StreamClient opens.

"v1" — WS /v1/stream/analyze, analyzed by the Inter-1 model.
"v2" — WS /v2/stream/analyze, analyzed by the Inter-2 model.

The two endpoints share one protocol: the same authentication, the same session.ready limits and session config, the same video framing, and the same server→client envelopes in the same order. Only the model behind the analysis differs, so the same client, handlers and types serve both.

StreamModel

type StreamModel = "inter-2" | "inter-2-audio" | "inter-2-deep";

Inter-2 models a WS /v2/stream/analyze session can select through its session config.

"inter-2" (the default) reads the video.
"inter-2-audio" hears the audio alone: an audio-only recording is accepted, the video of a recording that carries one is dropped, and raw PCM frames are accepted once an RawAudioFormat is declared.
"inter-2-deep" is an upload model; naming it on the stream is refused.

WS /v1/stream/analyze offers no selection and rejects any value.

PcmSampleRate

type PcmSampleRate = 16000 | 24000;

Sample rates accepted for raw PCM frames.

StreamEvent

type StreamEvent = 
  | SessionReadyEvent
  | SessionUpdatedEvent
  | SessionClosingEvent
  | SessionEndedEvent
  | SignalDetectedEvent
  | SignalUpdatedEvent
  | SignalEndedEvent
  | EngagementUpdatedEvent
  | ConversationQualityUpdatedEvent
  | FeedbackGeneratedEvent
  | CoverageDroppedEvent
  | CoverageDegradedEvent
  | StreamErrorEvent;

Discriminated union of every server→client envelope.

WebSocketFactory

type WebSocketFactory = (url, protocols?) => WebSocketLike;

Constructs a WebSocketLike from a URL and optional subprotocols.

Parameters

Parameter	Type
`url`	`string`
`protocols?`	`string` \| `string`[]

Returns

WebSocketLike

ScopeValue

type ScopeValue = 
  | "interhumanai.upload"
  | "interhumanai.stream"
  | "interhumanai.realtime"
  | "interhumanai.upload.inter-2"
  | "interhumanai.upload.inter-2-audio"
  | "interhumanai.upload.inter-2-deep"
  | "interhumanai.stream.inter-2"
  | "interhumanai.stream.inter-2-audio";

OAuth-style scopes an API key can carry. A scope names one operation, or one (operation, model) cell of the Inter-2 routes: the bare interhumanai.upload and interhumanai.stream are the Inter-1 cells, and interhumanai.<operation>.<model> is one model on the matching /v2 route. No scope implies another, and there is no interhumanai.stream.inter-2-deep: Deep is an upload-only model.

SignalType

type SignalType = 
  | "agreement"
  | "confidence"
  | "confusion"
  | "disagreement"
  | "disengagement"
  | "engagement"
  | "frustration"
  | "hesitation"
  | "interest"
  | "skepticism"
  | "stress"
  | "tension"
  | "uncertainty";

Social signal types the model can detect. tension and frustration name the same underlying negative state but keep the detecting source legible: tension comes only from the Realtime API’s visual track, while frustration comes from every other source (Inter-1, the realtime audio track, and the upload and stream endpoints).

Probability

type Probability = "high" | "medium" | "low";

Confidence level attached to a detected signal.

EngagementLevel

type EngagementLevel = "engaged" | "neutral" | "disengaged";

Engagement state of the analyzed subject.

GoalDimension

type GoalDimension = "clarity" | "authority" | "energy" | "rapport" | "learning";

Conversation-quality dimensions a caller can flag as a session goal.

IncludeFlag

type IncludeFlag = "conversation_quality_overall" | "conversation_quality_timeline";

Optional response sections the caller may request.

FeedbackType

type FeedbackType = "feedback" | "no_feedback";

Discriminator for the structured interaction-feedback result.

VideoInput

type VideoInput = 
  | Blob
  | {
  data: Uint8Array | ArrayBuffer | Blob;
  filename?: string;
  contentType?: string;
};

The video to analyze. Either a Blob/File (browsers, or Node 18+ where Blob is global) or raw bytes plus a filename.

Union Members

Blob

Type Literal

{
  data: Uint8Array | ArrayBuffer | Blob;
  filename?: string;
  contentType?: string;
}

data

data: Uint8Array | ArrayBuffer | Blob;

Raw video bytes.

filename?

optional filename?: string;

Filename to report in the multipart part, e.g. "clip.mp4".

contentType?

optional contentType?: string;

MIME type, e.g. "video/mp4".

UploadModel

type UploadModel = "inter-2" | "inter-2-audio" | "inter-2-deep";

Inter-2 models a POST /v2/upload/analyze job can name. inter-2-audio is served today. inter-2 and inter-2-deep are valid values the route does not serve yet; submitting them rejects with an InterhumanApiError whose errorId is ih4020.

UploadJobStatus

type UploadJobStatus = "queued" | "running" | "completed" | "failed";

Lifecycle of a POST /v2/upload/analyze job. queued and running are transient; completed and failed are terminal, and the job stays readable until its expires_at.

Variables

HTTP_BASE_URLS

const HTTP_BASE_URLS: Record<Environment, string>;

HTTP base URLs per named environment.

AnalysisGroup

AnalysisGroup: {
  Visual: "visual";
  Audio: "audio";
};

Named analysis-group constants.

Type Declaration

Visual

readonly Visual: "visual" = "visual";

The visual-signals track.

Audio

readonly Audio: "audio" = "audio";

The audio tracks.

RealtimeRecommendationFrequency

RealtimeRecommendationFrequency: {
  High: "high";
  Medium: "medium";
  Low: "low";
};

Named recommendation-frequency constants.

Type Declaration

High

readonly High: "high" = "high";

A recommendation runs every 10 seconds of analyzed video.

Medium

readonly Medium: "medium" = "medium";

A recommendation runs every 20 seconds of analyzed video.

Low

readonly Low: "low" = "low";

A recommendation runs every 30 seconds of analyzed video.

SDK_NAME

const SDK_NAME: "typescript" = "typescript";

This SDK’s canonical name in API telemetry.

SDK_VERSION

const SDK_VERSION: string = packageJson.version;

This SDK’s package version, read from package.json at build time so there is no second version literal to keep in sync.

SDK_HEADER_NAME

const SDK_HEADER_NAME: "X-Interhuman-SDK" = "X-Interhuman-SDK";

HTTP header carrying the SDK identity.

STREAM_ENDPOINT_PATHS

const STREAM_ENDPOINT_PATHS: Readonly<Record<StreamApiVersion, string>>;

Endpoint path opened for each StreamApiVersion.

StreamModel

StreamModel: {
  Inter2: "inter-2";
  Inter2Audio: "inter-2-audio";
  Inter2Deep: "inter-2-deep";
};

Named stream-model constants.

Type Declaration

Inter2

readonly Inter2: "inter-2" = "inter-2";

Inter2Audio

readonly Inter2Audio: "inter-2-audio" = "inter-2-audio";

Inter2Deep

readonly Inter2Deep: "inter-2-deep" = "inter-2-deep";

WS_READY_STATE

const WS_READY_STATE: {
  CONNECTING: 0;
  OPEN: 1;
  CLOSING: 2;
  CLOSED: 3;
};

readyState constants, mirrored from the WHATWG WebSocket spec.

Type Declaration

CONNECTING

readonly CONNECTING: 0 = 0;

OPEN

readonly OPEN: 1 = 1;

CLOSING

readonly CLOSING: 2 = 2;

CLOSED

readonly CLOSED: 3 = 3;

Scope

const Scope: {
  Upload: "interhumanai.upload";
  Stream: "interhumanai.stream";
  Realtime: "interhumanai.realtime";
  UploadInter2: "interhumanai.upload.inter-2";
  UploadInter2Audio: "interhumanai.upload.inter-2-audio";
  UploadInter2Deep: "interhumanai.upload.inter-2-deep";
  StreamInter2: "interhumanai.stream.inter-2";
  StreamInter2Audio: "interhumanai.stream.inter-2-audio";
};

Named scope constants.

Type Declaration

Upload

readonly Upload: "interhumanai.upload" = "interhumanai.upload";

Grants access to POST /v1/upload/analyze (Inter-1).

Stream

readonly Stream: "interhumanai.stream" = "interhumanai.stream";

Grants access to WS /v1/stream/analyze (Inter-1).

Realtime

readonly Realtime: "interhumanai.realtime" = "interhumanai.realtime";

Grants access to WS /v0/realtime/analyze.

UploadInter2

readonly UploadInter2: "interhumanai.upload.inter-2" = "interhumanai.upload.inter-2";

Grants POST /v2/upload/analyze with model: "inter-2", and reading the account’s jobs.

UploadInter2Audio

readonly UploadInter2Audio: "interhumanai.upload.inter-2-audio" = "interhumanai.upload.inter-2-audio";

Grants POST /v2/upload/analyze with model: "inter-2-audio", and reading the account’s jobs.

UploadInter2Deep

readonly UploadInter2Deep: "interhumanai.upload.inter-2-deep" = "interhumanai.upload.inter-2-deep";

Grants POST /v2/upload/analyze with model: "inter-2-deep", and reading the account’s jobs.

StreamInter2

readonly StreamInter2: "interhumanai.stream.inter-2" = "interhumanai.stream.inter-2";

Grants WS /v2/stream/analyze with model: "inter-2".

StreamInter2Audio

readonly StreamInter2Audio: "interhumanai.stream.inter-2-audio" = "interhumanai.stream.inter-2-audio";

Grants WS /v2/stream/analyze with model: "inter-2-audio".

GoalDimension

GoalDimension: {
  Clarity: "clarity";
  Authority: "authority";
  Energy: "energy";
  Rapport: "rapport";
  Learning: "learning";
};

Named goal-dimension constants.

Type Declaration

Clarity

readonly Clarity: "clarity" = "clarity";

Authority

readonly Authority: "authority" = "authority";

Energy

readonly Energy: "energy" = "energy";

Rapport

readonly Rapport: "rapport" = "rapport";

Learning

readonly Learning: "learning" = "learning";

IncludeFlag

IncludeFlag: {
  ConversationQualityOverall: "conversation_quality_overall";
  ConversationQualityTimeline: "conversation_quality_timeline";
};

Named include-flag constants.

Type Declaration

ConversationQualityOverall

readonly ConversationQualityOverall: "conversation_quality_overall" = "conversation_quality_overall";

ConversationQualityTimeline

readonly ConversationQualityTimeline: "conversation_quality_timeline" = "conversation_quality_timeline";

UploadModel

UploadModel: {
  Inter2: "inter-2";
  Inter2Audio: "inter-2-audio";
  Inter2Deep: "inter-2-deep";
};

Named upload-model constants.

Type Declaration

Inter2

readonly Inter2: "inter-2" = "inter-2";

Inter2Audio

readonly Inter2Audio: "inter-2-audio" = "inter-2-audio";

Inter2Deep

readonly Inter2Deep: "inter-2-deep" = "inter-2-deep";

DEFAULT_JOB_POLL_INTERVAL_MS

const DEFAULT_JOB_POLL_INTERVAL_MS: 2000 = 2000;

Default interval between status reads in UploadClient.waitForJob.

Functions

resolveHttpBaseUrl()

function resolveHttpBaseUrl(options): string;

Resolve the HTTP base URL for a client. baseUrl takes precedence (use it to point at a local server, e.g. http://localhost:8080); otherwise the named environment is used, defaulting to production. The returned URL never has a trailing slash.

Parameters

Parameter	Type
`options`	{ `baseUrl?`: `string`; `environment?`: `Environment`; }
`options.baseUrl?`	`string`
`options.environment?`	`Environment`

Returns

string

httpToWsBaseUrl()

function httpToWsBaseUrl(httpBaseUrl): string;

Derive the WebSocket origin from an HTTP base URL: https→wss, http→ws. Any path on the base URL is preserved (so a proxy mount point survives), with the trailing slash trimmed.

Parameters

Parameter	Type
`httpBaseUrl`	`string`

Returns

string

sdkHeaderValue()

function sdkHeaderValue(): string;

The X-Interhuman-SDK value this SDK sends: typescript/<version>.

Returns

string

isTerminal()

function isTerminal(job): boolean;

Whether a job has finished, successfully or not.

Parameters

Parameter	Type
`job`	`UploadJob`

Returns

boolean

​Classes

​AuthClient

​Constructors

Constructor

Parameters

Returns

​Methods

createToken()

Parameters

Returns

Throws

createClientToken()

Parameters

Returns

Throws

revokeClientToken()

Parameters

Returns

Throws

​StaticTokenProvider

​Implements

​Constructors

Constructor

Parameters

Returns

​Methods

getToken()

Returns

Implementation of

​TokenManager

​Implements

​Constructors

Constructor

Parameters

Returns

​Methods

getToken()

Returns

Implementation of

invalidate()

Returns

​InterhumanClient

​Example

​Constructors

Constructor

Parameters

Returns

​Properties

​Methods

stream()

Parameters

Returns

realtime()

Returns

getToken()

Returns

​InterhumanError

​Extends

​Extended by

​Constructors

Constructor

Parameters

Returns

Overrides

​InterhumanApiError

​Extends

​Constructors

Constructor

Parameters

Returns

Overrides

​Properties

​UploadJobTimeoutError

​Extends

​Constructors

Constructor

Parameters

Returns

Overrides

​Properties

Classes

AuthClient

Constructors

Methods

StaticTokenProvider

Implements

Constructors

Methods

TokenManager

Implements

Constructors

Methods

InterhumanClient

Example

Constructors

Properties

Methods

InterhumanError

Extends

Extended by

Constructors

InterhumanApiError

Extends

Constructors

Properties

UploadJobTimeoutError

Extends

Constructors

Properties

InterhumanConfigError

Extends

Constructors

RealtimeClient

Example

Extends

Constructors

Properties

Accessors

Methods