@alfe.ai/agent-api-client 0.3.0 → 0.4.0

package/dist/index.js

@@ -595,6 +595,47 @@ var AgentApiClient = class {
 			body: JSON.stringify(data)
 		});
 	}
+	/** Update the agent's own name and/or voice config. Returns the updated agent. */
+	async updateSelf(update) {
+		return this.request("/agent/self", {
+			method: "PATCH",
+			body: JSON.stringify(update)
+		});
+	}
+	/**
+	* Generate the agent's own avatar from a text prompt. The image is generated,
+	* stored, and set on the agent server-side; returns the updated agent.
+	*/
+	async generateAvatar(args) {
+		return this.request("/agent/avatar/generate", {
+			method: "POST",
+			body: JSON.stringify(args)
+		});
+	}
+	/**
+	* Get a presigned PUT URL to upload a new avatar image. Upload the bytes to
+	* `uploadUrl`, then call `finalizeAvatar(s3Key)` to set it on the agent.
+	*/
+	async presignAvatar(args) {
+		return this.request("/agent/avatar/presign", {
+			method: "POST",
+			body: JSON.stringify(args)
+		});
+	}
+	/**
+	* Finalize an avatar upload — validates ownership + size, then sets the
+	* agent's `avatarUrl` server-side. Returns the updated agent.
+	*/
+	async finalizeAvatar(s3Key) {
+		return this.request("/agent/avatar", {
+			method: "POST",
+			body: JSON.stringify({ s3Key })
+		});
+	}
+	/** List the platform voice catalogue (ElevenLabs) so the agent can pick its own voice. */
+	async listVoices() {
+		return this.request("/agent/voices");
+	}
 	/**
 	* Mint a fresh AES-256 data key for a specific (secret, field) pair. The
 	* encryption context is rebuilt server-side from `auth.tenantId` + the body
@@ -990,6 +1031,104 @@ var AgentApiClient = class {
 			body: JSON.stringify(entry)
 		}).catch(() => {});
 	}
+	async requestBrowserTakeover(args) {
+		return this.request("/agent/remote/takeover", {
+			method: "POST",
+			body: JSON.stringify(args)
+		});
+	}
+	async getRemoteSession(sessionId) {
+		return this.request(`/agent/remote/sessions/${encodeURIComponent(sessionId)}`);
+	}
+	async completeRemoteSession(sessionId) {
+		return this.request(`/agent/remote/sessions/${encodeURIComponent(sessionId)}/complete`, {
+			method: "POST",
+			body: JSON.stringify({})
+		});
+	}
+	/**
+	* Issue a request that returns the raw `Response` (no JSON parsing, no
+	* forced Content-Type). The caller sets `Content-Type`/`Accept` on `headers`
+	* and reads the body itself (`arrayBuffer()` / `json()`).
+	*
+	* A single retry fires only on the same transient statuses `request()`
+	* retries (authorizer-timeout 500 + LB 502/503/504) and transient network
+	* errors — before the route handler runs — so re-issuing a POST does not
+	* risk a duplicate side effect. Voice TTS/STT are effectively idempotent
+	* (re-synthesize / re-transcribe) and meter server-side keyed on the
+	* gateway requestId, so a retried transcription doesn't double-bill.
+	*/
+	async rawFetch(path, init) {
+		const url = `${this.apiUrl}${path}`;
+		init.headers.set("Authorization", `Bearer ${this.apiKey}`);
+		let lastError;
+		for (let attempt = 1; attempt <= 2; attempt++) try {
+			const res = await fetch(url, {
+				method: init.method,
+				headers: init.headers,
+				body: init.body,
+				signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS)
+			});
+			if (!res.ok) {
+				await res.text();
+				const error = /* @__PURE__ */ new Error(`Agent API request failed (${String(res.status)})`);
+				if (attempt === 1 && RETRYABLE_STATUS.has(res.status)) {
+					lastError = error;
+					await sleep(RETRY_DELAY_MS);
+					continue;
+				}
+				throw error;
+			}
+			return res;
+		} catch (err) {
+			if (attempt === 1 && isRetryableNetworkError(err)) {
+				lastError = err;
+				await sleep(RETRY_DELAY_MS);
+				continue;
+			}
+			throw err;
+		}
+		throw lastError instanceof Error ? lastError : new Error(String(lastError));
+	}
+	/**
+	* Text-to-speech. Returns raw PCM audio bytes plus their framing — the
+	* voice service defaults to 24 kHz / mono / 16-bit. Wrap in a WAV container
+	* to produce a playable file. Metered per character against the tenant
+	* credit pool server-side; TTS completes regardless of metering outcome.
+	*/
+	async tts(args) {
+		const headers = new Headers();
+		headers.set("Content-Type", "application/json");
+		headers.set("Accept", "audio/pcm");
+		const res = await this.rawFetch("/voice/tts", {
+			method: "POST",
+			headers,
+			body: JSON.stringify(args)
+		});
+		return {
+			audio: Buffer.from(await res.arrayBuffer()),
+			sampleRate: parseInt(res.headers.get("x-sample-rate") ?? "24000", 10),
+			channels: parseInt(res.headers.get("x-channels") ?? "1", 10),
+			bitDepth: parseInt(res.headers.get("x-bit-depth") ?? "16", 10)
+		};
+	}
+	/**
+	* Speech-to-text. Accepts raw linear16 (16-bit LE) mono PCM — NOT a WAV or
+	* other container (the endpoint transcribes with a fixed linear16 encoding,
+	* so a container header would be transcribed as noise). Strip any WAV header
+	* and pass `sampleRate` from it before calling. Metered by transcribed
+	* duration against the tenant credit pool server-side.
+	*/
+	async stt(args) {
+		const headers = new Headers();
+		headers.set("Content-Type", "application/octet-stream");
+		headers.set("x-sample-rate", String(args.sampleRate));
+		return (await (await this.rawFetch("/voice/stt", {
+			method: "POST",
+			headers,
+			body: args.audio
+		})).json()).data;
+	}
 };
 //#endregion
 export { AgentApiClient };