Overview

We've worked to make migration from OpenAI to Telnyx as painless as possible.

You will only need to:

1. Set the OPENAI_BASE_URL and OPENAI_API_KEY environment variables
2. Use an open-source model supported by Telnyx

Set the environment variables...

export OPENAI_BASE_URL='https://api.telnyx.com/v2/ai'
export OPENAI_API_KEY='KEY***'

and this code will work!

from openai import OpenAI

client = OpenAI()
chat_completion = client.chat.completions.create(
    # model="gpt-3.5-turbo",
    model="meta-llama/Meta-Llama-3-70B-Instruct",
    messages=[
        {"role": "user", "content": "Tell me about Telnyx"}
    ],
    temperature=0.0,
    stream=True,
)

If you'd like to use different environment variables, you may also pass these fields in the client constructor

from openai import OpenAI

client = OpenAI(
    api_key=os.getenv("TELNYX_API_KEY"),
    base_url=os.getenv("TELNYX_BASE_URL")
)
chat_completion = client.chat.completions.create(
    # model="gpt-3.5-turbo",
    model="meta-llama/Meta-Llama-3-70B-Instruct",
    messages=[
        {"role": "user", "content": "Tell me about Telnyx"}
    ],
    temperature=0.0,
    stream=True,
)

Telnyx supports the vast majority of parameters for Chat and Audio (and a few helpful ones that OpenAI does not).

See the full compatibility matrix below.

Chat Completions

Parameter	Description	Telnyx	OpenAI
messages	Provides chat context	✅	✅
model	Adjusts speed + quality	✅	✅
stream	Streams response	✅	✅
max_tokens	Limits output length	✅	✅
temperature	Adjusts predictability	✅	✅
top_p	Adjusts variety	✅	✅
frequency_penalty	Decreases repetition	✅	✅
presence_penalty	Decreases repetition...	✅	✅
n	Returns n responses	✅	✅
stop	Forces model to stop	✅	✅
logit_bias	Tweaks odds of results	✅	✅
logprobs	Returns odds of outputs	✅	✅
top_logprobs	-> For how many candidates?	✅	✅
seed	Reduces randomness	✅	✅
response_format	Ensures syntax (e.g. JSON)	✅	✅
guided_json	Ensures output conforms to schema	✅	❌
guided_regex	Ensures output conforms to regex	✅	❌
guided_choice	Ensures output conforms to choice	✅	❌
min_p	top_p alternative	✅	❌
use_beam_search	Explores more options	✅	❌
best_of	-> How many options?	✅	❌
length_penalty	-> Are long options bad?	✅	❌
early_stopping	-> How hard should it try?	✅	❌
tools	Helps model respond	✅	✅
functions	-> Outputs JSON for your code	🔜	✅
retrieval	-> Uses your docs (e.g. PDFs)	✅	❌
tool_choice	How does model choose?	🔜	✅
user	Tracks users	❌	✅

Transcriptions (BETA)

Parameter	What does this do?	Telnyx	OpenAI
file	Provides audio data	✅	✅
model	Adjusts speed + quality	✅	✅
response_format	Adjusts output format	✅	✅
timestamp_granularities[]	Adds timestamps	✅	✅
-> segment	-> per audio segment	✅	✅
-> word	-> per word	❌	✅
language	Improves accuracy	❌	✅
prompt	Guides style	❌	✅
temperature	Adjusts "creativity"	❌	✅