Repository navigation
Expand file tree
/
Copy path.env.example
More file actions
72 lines (65 loc) · 3.44 KB
/
Copy path.env.example
File metadata and controls
72 lines (65 loc) · 3.44 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
# Copy this file to .env and paste your key below.
#
# cp .env.example .env
#
# Get a free key at https://console.deepgram.com/signup?jump=keys
#
# This is the only credential Steps 1-8 need. The agent thinks with OpenAI's
# gpt-4o-mini, but Deepgram brokers that call -- no OpenAI account, no second
# key, no second bill. (Step 6b is the optional exception, at the bottom.)
DEEPGRAM_API_KEY=
# Which Deepgram hosting location to use. Optional -- leave it blank and every
# step talks to the global endpoint, which is what most rooms want.
#
# global api.deepgram.com + agent.deepgram.com (default)
# eu api.eu.deepgram.com processed inside the EU
# au api.au.deepgram.com processed inside Australia
# in api.in.deepgram.com processed inside India
#
# Your key works in every region; only the base URL changes. Setting it here
# moves all eight steps at once -- no step's code names a region.
#
# Run Step 1 after changing it. It starts the workshop's agent against the
# endpoint you picked and tells you whether all three models are served there,
# which is the question that actually differs between regions. See
# web/region.py for the details.
DEEPGRAM_REGION=
# ---------------------------------------------------------------------------
# Optional -- Step 6b only. Leave this whole block empty and every step still
# runs, Step 6b included; it falls back to OpenAI when it finds no credentials.
#
# Step 6b moves the agent's brain to Amazon Bedrock, in your own AWS account.
# Deepgram does not broker Bedrock the way it brokers OpenAI, so this is the one
# place in the workshop where a second credential and a second bill appear.
#
# Scope this credential tightly. It travels to Deepgram inside the Settings
# message, because Deepgram makes the Bedrock call as you. An IAM user limited
# to bedrock:InvokeModelWithResponseStream on the one model ARN you are using is
# the right size; your admin keys are not.
#
# Note for anyone arriving from the Pipecat edition of this workshop:
# AWS_BEARER_TOKEN_BEDROCK does NOT work here. That is a botocore convenience,
# and there is no botocore in this edition -- Deepgram takes an access key and
# secret, or STS credentials, and nothing else.
# ---------------------------------------------------------------------------
# Must match the region your Bedrock model access is granted in. It sets both
# the credentials' region and the endpoint URL, which have to agree.
# AWS_REGION=us-east-2
# Long-lived IAM keys, created in the IAM console (NOT the Bedrock console).
# AWS_ACCESS_KEY_ID=
# AWS_SECRET_ACCESS_KEY=
# Short-lived STS credentials instead? Set the pair above to the temporary ones
# and add the session token here. Its presence is what switches the credentials
# type from "iam" to "sts".
# AWS_SESSION_TOKEN=
# Optional. Defaults to zai.glm-4.7-flash -- fast and cheap, which matters more
# than raw capability when someone is waiting to hear a reply. Model IDs are
# passed through to Bedrock untouched, so nothing validates this locally; a typo
# comes back from AWS at handshake time.
# AWS_BEDROCK_MODEL=
# Bedrock model access is granted PER MODEL, PER REGION. Enable the model above
# for the region above at:
# https://console.aws.amazon.com/bedrock/home#/modelaccess
# Some model families (Anthropic's among them) additionally require a one-time
# use case form on that same page, which is not instant. Sort this out before
# the day, not on it.