2026/01/17

Thini-AI Voice Agent

System Architecture

Think-AI's Voice Agent API

Overview

This project provides a serverless API suite for LLM processing, Translation, Text-to-Speech (TTS), and Speech-to-Text (STT) services. It runs on AWS Lambda using a containerized Node.js environment and is exposed via AWS HTTP API Gateway.

1. API Specification

Base URL: https://api.60-think.com (or your API Gateway URL)
Authentication: API Key via Header x-api-key

Authentication

All requests must include the x-api-key header.

-H "x-api-key: <YOUR_CLIENT_API_KEY>"

Endpoints

POST /llm

Process text using various LLM providers (Google Gemini, DeepSeek).

Request:

{
  "task": "expand", // Optional: "expand", "headline"
  "prompt": "Explain Quantum Computing",
  "provider": "gemini", // "gemini", "deepseek", "openai"
  "model": "gemini-2.5-flash",
  "userId": "1" // For usage logging
}

Response:

{
  "result": "Quantum computing uses quantum bits...",
  "provider": "gemini",
  "model": "gemini-2.5-flash"
}

Curl Example:

curl -X POST https://api.60-think.com/llm \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_KEY" \
  -d '{"prompt":"Hello", "provider":"gemini", "userId":"1"}'

POST /translate

Translate text using AWS Translate.

Request:

{
  "text": "Hello, world",
  "targetLang": "es", // ISO code (ja, es, fr...)
  "userId": "1"
}

Response:
```
{
  "translated": "Hola, mundo"
}
```

POST /tts

Convert text to speech using AWS Polly.

Request:

{
  "text": "Hello world",
  "voice": "Joanna", // Optional
  "userId": "1"
}

Response:

{
  "url": "https://s3.ap-northeast-1.amazonaws.com/your-audio-bucket/tts-....mp3"
}

POST /stt/file

Transcribe audio files using AWS Transcribe.

Request:

{
  "fileUrl": "https://example.com/audio.mp3",
  "language": "en-US", // Optional, default en-US
  "userId": "1"
}

Response:

{
  "message": "Job Completed",
  "jobName": "stt-manual-170...",
  "transcript": "Hello world from audio..."
}

POST /stt/auth

Get temporary AWS credentials for real-time streaming transcription (e.g. from frontend).

Request:
```
{
  "userId": "1"
}
```

Response:

{
  "accessKeyId": "ASIA...",
  "secretAccessKey": "...",
  "sessionToken": "...",
  "region": "ap-northeast-1"
}

2. Deployment

Prerequisites

AWS CLI configured
Docker installed
.env.production file populated

Step 1: Deploy Lambdas (Infrastructure)

Run the setup script to create ECR repos, IAM roles, build the Docker image, and create/update Lambda functions.

./scripts/setup.sh

Step 2: Deploy API Gateway

Run the deploy API script to create or update the HTTP API, routes, integrations, and Authorizer.

./scripts/deploy-api.sh

Step 3: VPC Attachment (Optional/Security)

To attach Lambdas to the VPC (for internal logging privacy):

Warning: Requires NAT Gateway for external API access (LLM/Translate).
Run: ./scripts/attach-vpc.sh

Step 4: Update Configuration

To update environment variables without full redeployment:

./scripts/update-config.sh

3. Configuration & Environment Variables

Create a .env.production file with the following keys:

Variable	Description
`AWS_REGION`	AWS Region (default: ap-northeast-1)
`CLIENT_API_KEY`	Secret key clients must use in `x-api-key` header
`GEMINI_API_KEY`	API Key for Google Gemini
`DEEPSEEK_API_KEY`	API Key for DeepSeek
`AUDIO_BUCKET`	S3 bucket for TTS audio files
`TRANSCRIPTS_BUCKET`	S3 bucket for STT files
`GHOST_USAGE_API_URL`	URL for Ghost CMS Usage Logging
`GHOST_ADMIN_API_KEY`	Ghost Admin Key for logging auth

Notes

Security: The API uses a custom Lambda Authorizer (ai-authorizer) that validates the x-api-key header.
Logging: Usage stats are logged to Ghost CMS. If VPC is enabled, verify internal connectivity.

4. Remote API Verification Report (2026-01-18)

Summary

The ai-voice-suite API was tested against the production environment https://api.60-think.com on 2026-01-18.
The x-api-key authentication is functioning correctly.

Test Results

LLM Providers

Provider	Model	Status	Notes
DeepSeek	`deepseek-chat`	PASS	Response received successfully.
GLM	`GLM-4`	PASS	Response received successfully.
Gemini	`gemini-2.5-flash`	PASS	Response received successfully.
Gemini	`gemini-3-pro-preview`	PASS (Quota)	Authentication successful, but hit API rate limits (429).
Gemini	`gemini-2.5-flash-image` (Nano)	PASS (Quota)	Authentication successful, but hit API rate limits (429).
Gemini	`veo-3.0-generate-001`	FAIL	Model requires `predictLongRunning` method, which is not yet supported.

Other Endpoints

Endpoint	Function	Status	Notes
`/translate`	AWS Translate	PASS	Text translated successfully.
`/tts`	AWS Polly	PASS	Audio URL generated.
`/stt/file`	AWS Transcribe	PASS	Job started.
`/webhook`	Ghost Webhook	PASS	Webhook processed.

API Test

API Test Summary

Date: Sat Jan 17 22:30:30 JST 2026
API URL: https://api.60-think.com

Test: Unauthorized Access

Endpoint: /llm
Response (No Header):

{"message":"Unauthorized"}

Test: llm-deepseek

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Explain the concept of recursion.",
    "provider": "deepseek",
    "model": "deepseek-chat",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-preview:generateContent: fetch failed"}

Duration: 10945ms

Test: llm-nano

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Describe a sunset.",
    "provider": "gemini",
    "model": "gemini-2.5-flash-image",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30232ms

Test: tts

Endpoint: /tts
Payload:

{
    "text": "This is a test of the audio system.",
    "voice": "Joanna",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent: fetch failed"}

Duration: 11683ms

Test: llm-veo

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "A cat playing piano.",
    "provider": "gemini",
    "model": "veo-3.0-generate-001",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30175ms

Test: llm-gemini-flash

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "List 3 programming languages.",
    "provider": "gemini",
    "model": "gemini-2.5-flash",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/veo-3.0-generate-001:generateContent: fetch failed"}

Duration: 9989ms

Test: llm

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Tell me a joke about APIs",
    "provider": "gemini",
    "model": "gemini-2.5-flash",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent: fetch failed"}

Duration: 10800ms

Test: llm-gemini-pro

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Describe the future of AI.",
    "provider": "gemini",
    "model": "gemini-3-pro-preview",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent: fetch failed"}

Duration: 10028ms

Test: translate

Endpoint: /translate
Payload:

{
    "text": "Sixty years of experiences, countless stories. This is your exclusive platform to share life wisdom and relive wonderful legends.",
    "source": "en",
    "target": "ja",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30277ms

Test: stt

Endpoint: /stt/file
Payload:

{
    "fileUrl": "https://dnicugzydez8x.cloudfront.net/60-think-prd/2025/11/speech_20251105082319224.mp3",
    "language": "zh-CN",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-preview:generateContent: fetch failed"}

Duration: 11869ms

Test: llm-nano

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Describe a sunset.",
    "provider": "gemini",
    "model": "gemini-2.5-flash-image",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent: fetch failed"}

Duration: 10163ms

Test: llm-veo

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "A cat playing piano.",
    "provider": "gemini",
    "model": "veo-3.0-generate-001",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30553ms

Test: tts

Endpoint: /tts
Payload:

{
    "text": "This is a test of the audio system.",
    "voice": "Joanna",
    "userId": "1"
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/veo-3.0-generate-001:generateContent: fetch failed"}

Duration: 10160ms

Test: llm

Endpoint: /llm
Payload:

{
    "task": "expand",
    "prompt": "Tell me a joke about APIs",
    "provider": "gemini",
    "model": "gemini-2.5-flash",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30283ms

Test: webhook

Endpoint: /webhook
Payload:

{
    "post": {
        "current": {
            "title": "New Blog Post",
            "html": "<p>Content with audio need.</p>"
        }
    }
}

Response:

{"error":"[GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent: fetch failed"}

Duration: 11589ms

Test: translate

Endpoint: /translate
Payload:

{
    "text": "Sixty years of experiences, countless stories. This is your exclusive platform to share life wisdom and relive wonderful legends.",
    "source": "en",
    "target": "ja",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30135ms

Test: stt

Endpoint: /stt/file
Payload:

{
    "fileUrl": "https://dnicugzydez8x.cloudfront.net/60-think-prd/2025/11/speech_20251105082319224.mp3",
    "language": "zh-CN",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 28637ms

Response:

{"message":"Service Unavailable"}

Duration: 30276ms

Test: tts

Endpoint: /tts
Payload:

{
    "text": "This is a test of the audio system.",
    "voice": "Joanna",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30258ms

Test: webhook

Endpoint: /webhook
Payload:

{
    "post": {
        "current": {
            "title": "New Blog Post",
            "html": "<p>Content with audio need.</p>"
        }
    }
}

Response:

{"message":"Service Unavailable"}

Duration: 28637ms

Test: stt

Endpoint: /stt/file
Payload:

{
    "fileUrl": "https://dnicugzydez8x.cloudfront.net/60-think-prd/2025/11/speech_20251105082319224.mp3",
    "language": "zh-CN",
    "userId": "1"
}

Response:

{"message":"Service Unavailable"}

Duration: 30501ms

Response:

{"message":"Service Unavailable"}

Duration: 30316ms

Test: webhook

Endpoint: /webhook
Payload:

{
    "post": {
        "current": {
            "title": "New Blog Post",
            "html": "<p>Content with audio need.</p>"
        }
    }
}

Response:

{"message":"Service Unavailable"}