Cutting OpenAI TTS API Costs with Smart Audio Caching

How much does ChatNexus TTS cost?

Direct answer: ChatNexus uses OpenAI's <code>tts-1</code> model, which costs $0.015 per 1,000 characters. Most agent responses are 200–500 characters, so typical cost is $0.003–$0.008 per voice response.

Costs vary with response length. Longer knowledge base snippets cost more to synthesize. OpenAI pricing page lists the exact rates by model and region.

How does ChatNexus caching work?

Direct answer: ChatNexus hashes each response text with SHA-256 and stores the audio file in Redis for 7 days. If the same response is generated again (e.g., 'Sorry, I don't have an answer' appears frequently), ChatNexus serves the cached audio instead of calling OpenAI again.

The cache is keyed by sha256(text), so identical texts always hit the same cache entry. This is especially effective for common fallback messages, greetings, and pre-synthesized 'thinking' sounds. Cache hits save the full OpenAI API call cost.

What's the typical cost savings from caching?

Direct answer: Typical ChatNexus deployments see 30–60% cost reduction via caching. High-repeat-answer use cases (FAQ bots, customer service) can reach 70%+; low-repeat use cases (unique knowledge base queries) see 15–25%.

A support bot answering the same 50 FAQs all day will hit cache on ~60% of requests. A research bot generating unique queries will see lower rates. ChatNexus provides cost analytics in the dashboard so you can monitor your savings.

Do I need to set up caching myself?

Direct answer: No. ChatNexus caching is automatic and enabled by default. No configuration needed—you get cost savings immediately.

The cache is transparent. ChatNexus manages key generation, TTL, and eviction. You can view cache hit rates in your agent's analytics dashboard under the TTS Costs section.

What are pre-synthesized filler phrases and how do they save money?

Direct answer: Filler phrases like 'One moment…', 'Thinking…', and 'Let me check…' are synthesized once and reused across all agents. This saves money on responses that start with thinking animations while the agent queries the knowledge base.

When an agent is processing a knowledge base query, ChatNexus can play a pre-recorded 'thinking' sound while waiting. These fillers are cached globally and reused millions of times, eliminating redundant TTS API calls across the entire platform.

Deploy an AI agent on your site in minutes

Connect your knowledge base, embed on any domain, and start free — no credit card required.

Create Your AI Agent Free