Context router for voice agents
Not every turn needs your LLM
Fewer model calls per conversation.
Simple turns
Confirmations, greetings, consent speech, hold responses handled locally.
Complex turns
Reasoning, decision-making, ambiguous input. These get full context and go to your model as normal.
The result
Fewer LLM calls. Lower cost. Better output quality.
Works with your platform
Deepgram
Agent config - set the LLM provider to a custom endpoint - SLNG URL + key.
ElevenLabs
Agent settings - LLM - Custom LLM - set Server URL, Model ID, and key to SLNG.
Vapi
Assistant config - Custom LLM - set the endpoint to SLNG + your key.
Vocode
Agent config - set the LLM base URL to SLNG.
Bolna
Agent config - set the LLM endpoint to SLNG.
Voiceflow
Project settings - custom LLM - point the model endpoint at SLNG.
Any OpenAI-compatible stack
Set your OpenAI client's [base_url] to SLNG and use your SLNG key.
TTS path optimization
Lower TTS costs without changing your voice
Route your TTS through SLNG. Repeated sentences served from cache. The provider never gets called.
Full cost every time.
The greeting your agent has said ten thousand times costs the same as the first time it said it.
Cache hit. Provider skipped.
Same voice, no synthesis call, no provider charge. Cache miss = passes through on your key. Nothing else changes.
Start with your LLM endpoint
Your agent config has a custom LLM endpoint. Set the LLM endpoint to SLNG's URL + your SLNG key.
1"custom_llm": {
2 "url": "{region}.llm-router.slng.ai"
3}Add TTS when you're ready
Route your TTS provider through SLNG. Repeated sentences served from cache, provider never called.
1"tts": {
2 "url": "https://api.slng.ai/v1/tts",
3 "key": "YOUR_SLNG_KEY"
4}Compliance and security
Getting started
Live instantly. Results within 24hrs
Swap the URL in your agent config. We show you the numbers against your current LLM costs.
1. Get your SLNG key
Sign up at app.slng.ai. Get your API key.
2. Point to SLNG //
Set the LLM endpoint to SLNG's URL + your SLNG key. Add TTS the same way when you're ready.
3. See the results
Give it 24 hours. Your dashboard shows LLM cost, latency, and quality vs. baseline.
How does the Context router reduce cost?
The Context router evaluates each turn against the call context. Simple turns that don't need your model are handled locally. Only complex turns go to your LLM. Result: fewer model calls, lower cost, same conversation quality.
Does SLNG work with my agent platform?
If it has a custom LLM field or accepts an OpenAI-compatible endpoint, yes. ElevenLabs, Deepgram, Vapi, Vocode, Bolna, Voiceflow confirmed.
Does SLNG cache TTS audio?
Yes. When your agent repeats a sentence, SLNG serves the cached audio directly. The TTS provider never gets called. Same voice, no synthesis, no provider charge. Cache miss passes through on your key and gets cached for next time.
Do I need to change my TTS provider?
No. Route your TTS through SLNG. Your existing provider key, voice, and settings work unchanged.
Do I need to change my voice or STT provider?
No. SLNG only touches the LLM layer. Your voice, your STT, your platform stay the same.
How much does it cost?
US$ 0.0033 per agent minute. No contracts. No minimums. Your voice platform billing is separate.
Is my data safe?
PII redaction built in. ISO 27001 certified. HIPAA compliant. GDPR compliant. SOC 2 Type II in audit.
Stay in the loop
Unmuted.
Resources
Contact
Pricing
©2026 SLNG