[Example] 530 — Multi-Provider Chat Completions Proxy for Voice Agent (Python) by github-actions[bot] · Pull Request #210 · deepgram/examples

github-actions · 2026-04-08T08:59:03Z

New example: Multi-Provider Chat Completions Proxy for Voice Agent

Integration: Deepgram Voice Agent API | Language: Python | Products: agent

What this shows

A FastAPI proxy server that exposes an OpenAI-compatible /v1/chat/completions endpoint, routing requests to multiple LLM backends (OpenAI, AWS Bedrock). The Deepgram Voice Agent API uses this proxy via think.endpoint.url, letting developers swap LLM providers without changing application code.

The demo script connects a Deepgram Voice Agent to the proxy, demonstrating how STT (nova-3) and TTS (aura-2) run natively on Deepgram while LLM reasoning routes through a configurable backend.

Required secrets

DEEPGRAM_API_KEY — Deepgram console
OPENAI_API_KEY — OpenAI dashboard
AWS credentials (optional, for Bedrock provider): AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION

Tests

✅ Tests passed (8/8)

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

Built by Engineer on 2026-04-08

…oice Agent (Python)

github-actions · 2026-04-08T16:50:02Z

Code Review

Overall: CHANGES REQUESTED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed in 3.76s

All 8 tests pass — proxy routes to OpenAI successfully and Voice Agent accepts custom endpoint settings via real WebSocket connection.

Integration genuineness

❌ FAIL — NO RAW PROTOCOL CHECK (check 6)

demo_agent.py uses websockets.sync.client.connect() directly to wss://agent.deepgram.com/v1/agent/converse instead of the Deepgram Python SDK's agent.v1.connect method. The test (test_voice_agent_accepts_custom_endpoint_settings) does the same.

The deepgram-sdk==6.1.1 is listed in requirements.txt but is never imported in any source file. The SDK has full Voice Agent support via DeepgramClient.agent.v1.connect() — this should be used instead of raw WebSocket connections.

Other checks pass:

✅ Platform SDK is in requirements (deepgram-sdk==6.1.1, correct version)
✅ Real API calls to OpenAI and Deepgram Voice Agent (not mocked)
✅ .env.example lists both DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Tests exit(2) on missing credentials
✅ BYPASS CHECK: N/A — this is a proxy example; no partner audio SDK to route through

Code quality

Issues found:

Raw WebSocket instead of SDK — demo_agent.py and the integration test use websockets library directly. Must use deepgram-sdk's agent client.
Missing tags: ["deepgram-examples"] — Voice Agent Settings messages don't include the required tag. (Fixed locally: added to build_settings() in demo_agent.py and to the test's Settings dict.)
Unused json import in providers.py line 10. (Fixed locally: removed.)
Unused dependency — deepgram-sdk==6.1.1 is in requirements.txt but never imported anywhere in src/.
__pycache__ not gitignored — no .gitignore in the example directory.

What's good:

✅ No hardcoded credentials — all from env vars
✅ Good error handling (missing messages, invalid provider, provider errors)
✅ Tests import from src/ and test the actual proxy endpoints
✅ Server test spins up via TestClient and makes real HTTP requests
✅ No word-list assertions — only structural checks on responses
✅ Credential check runs first before imports
✅ SDK version is correct (deepgram-sdk==6.1.1)

Documentation

✅ README covers what you'll build, all env vars with links, run instructions
✅ .env.example present and complete
✅ Key parameters table is helpful

Fixes attempted

I fixed the tags and unused import locally but could not push (permission denied). The raw WebSocket → SDK migration is an architecture change (~30+ lines across demo_agent.py and the integration test) that should be handled by the fix agent.

Required changes:

Rewrite demo_agent.py to use DeepgramClient.agent.v1.connect() instead of raw websockets
Rewrite test_voice_agent_accepts_custom_endpoint_settings to use the SDK's agent client
Add tags=["deepgram-examples"] to the Settings (fixed locally, needs push)
Remove unused json import from providers.py (fixed locally, needs push)
Add a .gitignore with __pycache__/

Please address the items above. The fix agent will pick this up.

Review by Lead on 2026-04-08

… in 530-voice-agent-multi-provider-proxy-python - Rewrite demo_agent.py to use DeepgramClient.agent.v1.connect() and send_settings() instead of raw websockets - Rewrite integration test to use the SDK's agent client - Add tags=["deepgram-examples"] to Settings messages - Remove unused json import from providers.py - Remove direct websockets dependency (SDK provides it) - Add .gitignore with __pycache__/

github-actions · 2026-04-08T17:30:47Z

Fix applied

Root cause: demo_agent.py and the integration test used raw websockets instead of the Deepgram Python SDK's agent.v1.connect() method; also missing tags=["deepgram-examples"], unused imports, and no .gitignore.

Change: Rewrote demo_agent.py and test_voice_agent_accepts_custom_endpoint_settings to use DeepgramClient.agent.v1.connect() / send_settings() with typed AgentV1Settings. Added tags=["deepgram-examples"] to all Settings messages. Removed unused json import from providers.py. Removed direct websockets pin from requirements.txt (SDK provides it). Added .gitignore with __pycache__/.

Tests after fix ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed in 1.99s

✓ Fix verified — tests pass.

Fix by Lead on 2026-04-08

github-actions · 2026-04-09T00:57:26Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed, 2 warnings in 2.50s

All 8 tests pass with real credentials. OpenAI provider returns valid completions. Deepgram Voice Agent WebSocket accepts custom endpoint settings and returns SettingsApplied.

Integration genuineness

✅ Pass — all 6 checks clear:

Platform SDK imported — deepgram-sdk for Voice Agent API, httpx for OpenAI, boto3 for Bedrock
Real API calls — OpenAI completions via httpx.post(), Bedrock via client.converse(), Deepgram via client.agent.v1.connect()
.env.example lists real credentials — DEEPGRAM_API_KEY, OPENAI_API_KEY
Credential gate — Test file exits with code 2 if credentials missing (lines 16-18)
Bypass check — DeepgramClient usage is correct; it accesses the Voice Agent API (client.agent.v1.connect()), not raw STT/TTS. Audio flows through the Voice Agent's native STT (nova-3) and TTS (aura-2)
No raw protocol — No WebSocket(), ws(), or fetch() calls for Deepgram audio; all contact through the SDK

Code quality

✅ Official Deepgram SDK used — deepgram-sdk==6.1.1 (matches required version)
✅ tags=["deepgram-examples"] included in Voice Agent Settings (demo_agent.py:34, test_example.py:169)
✅ No hardcoded credentials — all sourced from environment
✅ Error handling — missing API keys raise RuntimeError, invalid providers return 400, provider errors return 502
✅ Tests import from src/ — proxy.app and demo_agent.build_settings tested via actual code paths
✅ Server tested properly — FastAPI TestClient makes real HTTP requests to all endpoints
✅ Real Deepgram API test — test_voice_agent_accepts_custom_endpoint_settings connects to agent.deepgram.com via SDK
✅ No word-list assertions — all assertions check response structure (keys, types, non-empty), not specific text
✅ Credential check runs first (top of test file, before any SDK imports)

Documentation

✅ README includes "What you'll build" section
✅ All env vars documented with where-to-get links in a table
✅ Install and run instructions provided
✅ "How it works" section explains the full flow
✅ .env.example present and complete

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-09

github-actions · 2026-04-09T05:06:22Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed, 2 warnings in 2.33s

All 8 tests passed with real credentials. OpenAI provider returned valid completions. Deepgram Voice Agent accepted custom endpoint settings and returned SettingsApplied.

Integration genuineness ✅

Platform SDK imported — deepgram-sdk==6.1.1 used via DeepgramClient and AgentV1Settings; OpenAI called via httpx as LLM backend; Bedrock via boto3
Real API calls — test_chat_completions_openai hits real OpenAI API; test_voice_agent_accepts_custom_endpoint_settings connects to real Deepgram Voice Agent WebSocket
.env.example lists real credentials — DEEPGRAM_API_KEY and OPENAI_API_KEY
Credential check exits 2 — test_example.py lines 15-18 exit with code 2 on missing creds before any SDK imports
BYPASS CHECK: PASS — Voice Agent used through client.agent.v1.connect() via the SDK. STT/TTS run natively on Deepgram. The proxy serves as the LLM backend — no bypass of the Deepgram interface
NO RAW PROTOCOL CHECK: PASS — No raw WebSocket() or fetch() for audio/transcription; SDK's agent connect method used

Code quality ✅

Official Deepgram SDK used (deepgram-sdk==6.1.1 — matches required version)
tags=["deepgram-examples"] present on Voice Agent Settings (demo_agent.py:35, test_example.py:138/169)
No hardcoded credentials
Error handling covers main failure cases (400 for bad provider/missing messages, 502 for provider errors, RuntimeError for missing API keys)
Tests import from src/ and call actual code (proxy.app, demo_agent.build_settings)
Server tests use TestClient to make real HTTP requests to FastAPI app
Assertions use response structure checks (choices, usage, status codes) — no word-list assertions
Credential check runs FIRST before SDK imports (test_example.py lines 1-18)

Documentation ✅

README includes "What you'll build" section
All env vars listed with where-to-get links
Install and run instructions present
.env.example present and complete
"How it works" section explains the architecture

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-09

github-actions · 2026-04-09T08:52:12Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.58s

All 8 tests pass with real credentials. The OpenAI provider returns valid chat completions, and the Deepgram Voice Agent WebSocket accepts custom think.endpoint settings (SettingsApplied received).

Integration genuineness

✅ Pass

Deepgram SDK (deepgram-sdk==6.1.1) is used for the Voice Agent connection via client.agent.v1.connect() — no bypass
OpenAI is called via httpx through the proxy's provider layer (real API call confirmed by test)
AWS Bedrock integration uses boto3.client("bedrock-runtime").converse() — real SDK call
No raw WebSocket or fetch calls for Deepgram API contact
Voice Agent settings include tags=["deepgram-examples"]
.env.example lists both DEEPGRAM_API_KEY and OPENAI_API_KEY
Credential check runs at module level (lines 6-18 of test_example.py) before any SDK imports

Code quality

✅ Official Deepgram SDK used (deepgram-sdk==6.1.1 — matches required version)
✅ tags=["deepgram-examples"] present in Voice Agent settings (demo_agent.py:34, test_example.py:169)
✅ No hardcoded credentials — all sourced from environment
✅ Error handling: provider errors → 502, validation errors → 400, missing credentials → RuntimeError
✅ Tests import from src/ (proxy.app, demo_agent.build_settings) and test actual code
✅ Server tests use FastAPI TestClient making real HTTP requests to endpoints
✅ Voice Agent test connects to the real Deepgram API and confirms SettingsApplied
✅ No word-list assertions — response structure checks only (non-deterministic safe)
✅ Credential check exits with code 2 before SDK imports

Documentation

✅ README includes "What you'll build" section, all env vars with where-to-get links, install/run instructions
✅ .env.example present and complete for required credentials
✅ Key parameters table documents Voice Agent configuration
✅ "How it works" section explains the full request flow

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-09

github-actions · 2026-04-09T12:45:48Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 1.80s

Real API calls confirmed:

OpenAI chat completion returned a valid response with token usage
Deepgram Voice Agent WebSocket connected, received Welcome + SettingsApplied with custom think.endpoint.url

Integration genuineness

✅ Pass — all 6 checks satisfied:

OpenAI SDK (httpx) and AWS Bedrock (boto3) imported and used for real API calls
test_chat_completions_openai makes a real OpenAI API call; test_voice_agent_accepts_custom_endpoint_settings connects to real Deepgram Voice Agent API
.env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY with source links
Credential check at top of test file exits with code 2 if missing
Deepgram Voice Agent accessed through official SDK (client.agent.v1.connect()); no bypass — STT/TTS run natively on Deepgram, LLM routes through the proxy as designed
No raw WebSocket or fetch calls to Deepgram; httpx.post() in providers.py targets OpenAI/Bedrock (not Deepgram)

Code quality

✅ deepgram-sdk==6.1.1 — correct required version
✅ tags=["deepgram-examples"] present in AgentV1Settings (demo_agent.py + test)
✅ No hardcoded credentials
✅ Error handling: missing messages (400), invalid provider (400), provider errors (502), missing API keys (RuntimeError)
✅ Tests import from src/ and exercise the actual proxy via FastAPI TestClient
✅ Server tests make real HTTP requests to the proxy endpoints
✅ Voice Agent test connects to real Deepgram WebSocket and asserts SettingsApplied
✅ No word-list assertions — checks response structure only (choices, usage, content length)
✅ Credential check runs first before any SDK imports

Documentation

✅ README covers what you'll build, all env vars with where-to-get links, install/run instructions, architecture explanation
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table and "How it works" flow diagram in README

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-09

github-actions · 2026-04-09T20:37:25Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.09s

All 8 tests passed with real credentials. OpenAI provider returned valid chat completions. Deepgram Voice Agent accepted custom endpoint settings (SettingsApplied received).

Integration genuineness

✅ Pass — all 6 checks clear:

Platform SDK imported: deepgram-sdk used via DeepgramClient for Voice Agent; httpx for OpenAI; boto3 for Bedrock
Real API calls: openai_completion() posts to OpenAI API; bedrock_completion() calls boto3.converse(); test connects to real Deepgram Voice Agent WebSocket
.env.example lists real credentials: DEEPGRAM_API_KEY, OPENAI_API_KEY
Credential check exits 2: test_example.py lines 16-18 exit with code 2 if credentials missing
Bypass check: Deepgram Voice Agent accessed through client.agent.v1.connect() SDK interface — not bypassed. Proxy sits between Voice Agent and LLM providers (not in the audio path)
No raw protocol: No raw WebSocket/fetch for Deepgram audio. SDK used throughout

Code quality

✅ deepgram-sdk==6.1.1 — matches required version
✅ tags=["deepgram-examples"] set in AgentV1Settings (demo_agent.py:35)
✅ No hardcoded credentials — all read from environment
✅ Error handling: 400 for missing messages/invalid provider, 502 for provider errors, RuntimeError for missing API keys
✅ Tests import from src/ (proxy.app, demo_agent.build_settings) — not standalone SDK tests
✅ Server tests use FastAPI TestClient for real HTTP requests to proxy endpoints
✅ Voice Agent test makes real WebSocket connection and verifies Welcome + SettingsApplied
✅ No word-list assertions — response checks are structural (keys present, non-empty, positive token count)
✅ Credential check runs before any SDK imports (lines 5-18 of test_example.py)

Documentation

✅ README includes "What you'll build", prerequisites, env vars table with links, install/run instructions, architecture walkthrough
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table documents Voice Agent settings

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-09

github-actions · 2026-04-10T01:04:54Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed, 2 warnings in 4.67s

All 8 tests pass with real credentials. OpenAI provider returns valid completions. Deepgram Voice Agent API accepts custom endpoint settings and returns SettingsApplied.

Integration genuineness

✅ Pass — all 6 checks satisfied:

Platform SDK imported: deepgram-sdk used via DeepgramClient and AgentV1Settings; httpx for OpenAI proxy calls; boto3 for Bedrock
Real API calls: test_chat_completions_openai makes a live OpenAI call through the proxy; test_voice_agent_accepts_custom_endpoint_settings connects to the real Deepgram Voice Agent WebSocket API
.env.example lists real credentials: DEEPGRAM_API_KEY and OPENAI_API_KEY both present
Credential guard: sys.exit(2) at top of test file before any SDK imports
Bypass check: DeepgramClient is used only via client.agent.v1.connect() — the Voice Agent SDK interface. No direct STT/TTS bypass; the proxy handles LLM routing, not audio
No raw protocol: No WebSocket(), ws(), or fetch() for Deepgram API contact

Code quality

✅ Official Deepgram SDK: deepgram-sdk==6.1.1 (matches required v6.1.1)
✅ tags=["deepgram-examples"] set in both demo_agent.py:build_settings() and in the Voice Agent test
✅ No hardcoded credentials — all secrets read from env
✅ Error handling: provider validation (400), missing messages (400), provider errors (502), missing credential RuntimeError
✅ Tests import from src/ — proxy.app via FastAPI TestClient and demo_agent.build_settings
✅ Server tested via real HTTP requests (FastAPI TestClient)
✅ No word-list assertions — only response structure checks (choices, usage, token count > 0)
✅ Credential check runs FIRST at module level before any src imports

Documentation

✅ README: "What you'll build" section, all env vars with where-to-get links, install/run instructions, architecture explanation
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table and "How it works" flow

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-10

github-actions · 2026-04-10T08:53:56Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed, 2 warnings in 2.15s

Integration genuineness

✅ Pass — All six checks passed:

Platform SDKs imported: httpx for OpenAI, boto3 for AWS Bedrock, deepgram-sdk for Voice Agent
Real API calls: openai_completion() posts to OpenAI API, bedrock_completion() calls Bedrock Converse, tests hit live endpoints
.env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY with source links
Credential check at module level with sys.exit(2) before SDK imports
Bypass check: DeepgramClient used correctly for Voice Agent WebSocket — the proxy routes LLM calls, not audio. Architecture is Voice Agent SDK → proxy → LLM providers
No raw WebSocket or fetch calls for Deepgram API contact

Code quality

✅ Official Deepgram SDK (deepgram-sdk==6.1.1) — correct version
✅ tags=["deepgram-examples"] present in Voice Agent Settings (demo_agent.py and test)
✅ No hardcoded credentials
✅ Error handling: missing messages (400), invalid provider (400), provider errors (502)
✅ Tests import from src/ (from proxy import app, from demo_agent import build_settings)
✅ Server tested via TestClient(app) with real HTTP requests to proxy + real API calls to OpenAI and Deepgram Voice Agent
✅ Assertions are structural (response shape, non-empty content, token count > 0) — no brittle word-list checks
✅ Credential check runs first at module level before any SDK imports

Documentation

✅ README covers what you'll build, prerequisites, all env vars with where-to-get links, install/run instructions, key parameters, architecture flow
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-10

github-actions · 2026-04-10T12:37:57Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.90s

All 8 tests pass including real OpenAI API calls through the proxy and a real Deepgram Voice Agent WebSocket connection with custom think.endpoint settings.

Integration genuineness

Pass — all 6 checks clear:

✅ OpenAI API called via httpx in providers.py; AWS Bedrock called via boto3.client("bedrock-runtime").converse()
✅ Real API calls — test_chat_completions_openai hits live OpenAI; test_voice_agent_accepts_custom_endpoint_settings opens a real Deepgram Voice Agent WebSocket
✅ .env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Credential check exits with code 2 before any imports (test_example.py:17-18)
✅ Bypass check N/A — this is a proxy sitting between Voice Agent and LLM backends; Deepgram STT/TTS flows through the Voice Agent API via client.agent.v1.connect() (SDK)
✅ No raw WebSocket/fetch for Deepgram audio — all Deepgram contact uses the official SDK

Code quality

✅ deepgram-sdk==6.1.1 — matches required Python SDK version
✅ tags=["deepgram-examples"] present in demo_agent.py:34 and test settings
✅ No hardcoded credentials; secrets read from env vars
✅ Error handling: proxy returns 400 for bad provider/missing messages, 502 for provider errors
✅ Tests import from src/ and test the actual proxy via TestClient(app) — not standalone SDK tests
✅ Server tests make real HTTP requests through FastAPI's TestClient
✅ Voice Agent test uses SDK's agent.v1.connect() to verify SettingsApplied
✅ No word-list assertions — response checks are structural (keys, types, non-empty)
✅ Credential check runs first at module level before SDK imports

Documentation

✅ README includes "What you'll build", all env vars with where-to-get links, install/run instructions, architecture explanation
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table documents Voice Agent config options

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-10

github-actions · 2026-04-10T16:40:27Z

Code Review

Overall: CHANGES REQUESTED

Tests ran ✅ (7/8) / ❌ (1/8)

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings FAILED

Failure: test_voice_agent_accepts_custom_endpoint_settings — The Deepgram Voice Agent API returns "Failed to resolve endpoint: https://api.openai.com/v1/chat/completions". The think.endpoint.url should be the base URL (https://api.openai.com/v1), not the full chat completions path. Confirmed that changing the URL to https://api.openai.com/v1 makes all 8 tests pass.

Fix: In tests/test_example.py line 179, change:

"url": "https://api.openai.com/v1/chat/completions",

to:

"url": "https://api.openai.com/v1",

(Fix was committed locally but could not be pushed due to permissions.)

Integration genuineness

✅ Pass — All 6 checks passed:

✅ Deepgram SDK imported and used (deepgram-sdk==6.1.1, client.agent.v1.connect())
✅ Real API calls to OpenAI (via httpx.post) and Deepgram Voice Agent (via SDK WebSocket)
✅ .env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Test exits with code 2 if credentials missing (credential check at top of test file)
✅ Bypass check pass — This is a proxy server; DeepgramClient is used properly through the Voice Agent SDK (client.agent.v1.connect()), not bypassing any partner interface
✅ No raw protocol — Uses Deepgram SDK for Voice Agent connection, no raw WebSocket/fetch for Deepgram API

Code quality

✅ Official Deepgram SDK used (deepgram-sdk==6.1.1 — matches required version)
✅ tags=["deepgram-examples"] present in demo_agent.py:build_settings() and in test settings
✅ No hardcoded credentials
✅ Error handling: provider validation, missing messages, credential checks, 502 on provider errors
✅ Tests import from src/ and test actual proxy code via TestClient(app)
✅ Server tests use FastAPI TestClient for real HTTP requests
✅ No specific word-list assertions — response structure checks only
✅ Credential check runs FIRST before any SDK imports (lines 5-18 of test file)

Documentation

✅ README includes: what you'll build, all env vars with where-to-get links, install/run instructions
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table and how-it-works section

Please apply the one-line fix to tests/test_example.py:179 — change the Voice Agent think.endpoint.url from https://api.openai.com/v1/chat/completions to https://api.openai.com/v1. The fix agent will pick this up.

Review by Lead on 2026-04-10

…ti-provider-proxy-python 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

github-actions · 2026-04-10T17:15:18Z

Fix applied

Root cause: The think.endpoint.url in the Voice Agent test used the full chat completions path (https://api.openai.com/v1/chat/completions) instead of the base URL (https://api.openai.com/v1), causing the Deepgram Voice Agent API to fail with "Failed to resolve endpoint".

Change: Changed tests/test_example.py:179 — think.endpoint.url from https://api.openai.com/v1/chat/completions to https://api.openai.com/v1.

Tests after fix ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED
======================== 8 passed, 2 warnings in 2.41s =========================

✓ Fix verified — tests pass.

Fix by Lead on 2026-04-10

github-actions · 2026-04-10T20:33:52Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.54s

All 8 tests pass with real credentials. OpenAI provider returns valid chat completion responses. Voice Agent WebSocket connection to Deepgram succeeds and receives SettingsApplied confirmation.

Integration genuineness

Pass — all 6 checks clear:

Platform SDK imported — deepgram-sdk for Voice Agent, httpx for OpenAI API, boto3 for AWS Bedrock
Real API calls — OpenAI chat completions via httpx.post() in providers.py:30-40; Bedrock Converse via boto3 in providers.py:55-80; Deepgram Voice Agent via client.agent.v1.connect() in demo_agent.py:92
.env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY
Credential guard — test_example.py:16-18 exits with code 2 on missing creds before any imports
Bypass check — DeepgramClient is used only for client.agent.v1.connect() (the Voice Agent WebSocket), which is the correct SDK path. LLM calls route through the proxy as intended.
No raw protocol — No WebSocket() or fetch() for Deepgram audio; SDK handles the connection

Code quality

deepgram-sdk==6.1.1 — correct pinned version
tags=["deepgram-examples"] present in Voice Agent settings (demo_agent.py:35)
No hardcoded credentials anywhere
Error handling: provider not found (400), missing messages (400), provider failures (502), missing API keys (RuntimeError)
Tests import from src/ via path insertion and test the actual proxy endpoints with FastAPI TestClient
Real Deepgram Voice Agent WebSocket test confirms SettingsApplied with custom think.endpoint
No word-list assertions — structure-only checks on non-deterministic LLM output
Proxy server correctly handles provider routing via env var and X-LLM-Provider header override

Documentation

README covers: what you'll build, prerequisites, all env vars with where-to-get links, install/run instructions, architecture walkthrough
.env.example present and complete with comment links to credential sources

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-10

github-actions · 2026-04-11T01:00:27Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.76s

All 8 tests passed with real credentials. OpenAI provider returned valid completions, Voice Agent WebSocket accepted custom endpoint settings (SettingsApplied received).

Integration genuineness

✅ Pass — all 6 checks satisfied:

Deepgram SDK (deepgram-sdk==6.1.1) imported and used via client.agent.v1.connect() for Voice Agent WebSocket; OpenAI called via httpx; Bedrock via boto3.client.converse()
Real API calls to OpenAI (test_chat_completions_openai) and Deepgram Voice Agent (test_voice_agent_accepts_custom_endpoint_settings) confirmed
.env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY with source links
Credential check at top of test file exits with code 2 if missing — no fake passes
Bypass check: N/A — this is a proxy server example; Deepgram Voice Agent API is used correctly via the SDK's agent.v1 interface. Audio flows through Deepgram's Voice Agent (STT/TTS), LLM calls flow through the proxy. No bypass.
No raw WebSocket(), ws(), or fetch() for audio/transcription. SDK used throughout.

Code quality

✅ Official Deepgram SDK used (deepgram-sdk==6.1.1 — matches required version)
✅ tags=["deepgram-examples"] included in Voice Agent settings (demo_agent.py:35)
✅ No hardcoded credentials — all read from environment
✅ Error handling: proxy returns 400 for missing messages/invalid provider, 502 for provider errors; providers raise on missing API keys
✅ Tests import from src/ (from proxy import app, from demo_agent import build_settings) — test the actual example code
✅ Server tests use FastAPI TestClient to make real HTTP requests to endpoints
✅ Voice Agent test connects to real Deepgram WebSocket and confirms SettingsApplied
✅ No word-list assertions — structural checks only (response shape, token counts, status codes)
✅ Credential check runs FIRST (lines 5-18 of test file) before any SDK imports

Documentation

✅ README includes "What you'll build" section
✅ All env vars documented with where-to-get links in table format
✅ Install and run instructions provided
✅ .env.example present with both required keys and source comments
✅ "How it works" section explains the full request flow

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-11

github-actions · 2026-04-11T08:31:07Z

@deepgram/devrel — VP escalation: this PR has status:review-passed but no E2E check has ever run. The lead-e2e workflow does not support workflow_dispatch, so it cannot be manually re-triggered.

State: labels=type:example,status:review-passed,language:python,integration:voice-agent-multi-provider-proxy; no checks have run.

Last activity: 2026-04-11T01:00:36Z

github-actions · 2026-04-11T08:33:31Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED                       [ 12%]
tests/test_example.py::test_models_endpoint PASSED                       [ 25%]
tests/test_example.py::test_chat_completions_missing_messages PASSED     [ 37%]
tests/test_example.py::test_chat_completions_invalid_provider PASSED     [ 50%]
tests/test_example.py::test_chat_completions_openai PASSED               [ 62%]
tests/test_example.py::test_provider_header_override PASSED              [ 75%]
tests/test_example.py::test_voice_agent_settings_builder PASSED          [ 87%]
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED [100%]

8 passed, 2 warnings in 2.69s

All 8 tests pass with real credentials. OpenAI provider returns valid completions. Deepgram Voice Agent WebSocket accepts custom think.endpoint settings and returns SettingsApplied.

Integration genuineness ✅

Platform SDK imported — deepgram-sdk used for Voice Agent WebSocket; httpx for OpenAI; boto3 for AWS Bedrock
Real API calls — openai_completion() makes real POST to OpenAI; bedrock_completion() makes real boto3 converse() call; test_voice_agent_accepts_custom_endpoint_settings connects to the real Deepgram Agent API
.env.example lists real credentials — DEEPGRAM_API_KEY, OPENAI_API_KEY
Credential check exits 2 — Top of test file checks .env.example vars and calls sys.exit(2) if missing
Bypass check: PASS — DeepgramClient is used only for the Voice Agent WebSocket (client.agent.v1.connect()), which is the correct integration path. Audio goes through the Agent API, LLM goes through the proxy — this is the intended architecture for think.endpoint.url
No raw protocol: PASS — No raw WebSocket() or fetch() for Deepgram API contact; SDK is used throughout

Code quality ✅

Official Deepgram SDK used (deepgram-sdk==6.1.1 — matches required version)
tags=["deepgram-examples"] present on Voice Agent settings (demo_agent.py:35)
No hardcoded credentials
Error handling: missing API keys raise RuntimeError, invalid providers return 400, provider errors return 502
Tests import from src/ — proxy.app via TestClient, demo_agent.build_settings
Server tests use FastAPI TestClient for real HTTP requests; Voice Agent test uses real WebSocket connection
No word-list assertions — structure-based checks only (response keys, token count > 0, non-empty content)
Credential check runs FIRST at top of test file before any SDK imports

Documentation ✅

README: "What you'll build" section, all env vars with where-to-get links, install/run instructions
.env.example present and complete
Key parameters table documents Voice Agent settings
"How it works" section explains the full request flow

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-11

github-actions · 2026-04-11T12:31:58Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.32s

All 8 tests pass with real credentials. OpenAI provider returns valid chat completions; Voice Agent WebSocket connects and receives SettingsApplied.

Integration genuineness

✅ Pass — all six checks satisfied:

Deepgram SDK (deepgram-sdk==6.1.1) and partner SDKs (httpx for OpenAI, boto3 for Bedrock) are imported and used
Real API calls: proxy forwards to OpenAI via httpx.post, Bedrock via boto3.client.converse(), and demo connects to Deepgram Voice Agent via client.agent.v1.connect()
.env.example lists DEEPGRAM_API_KEY and OPENAI_API_KEY with where-to-get comments
Tests exit with code 2 when credentials are missing (credential check at top of test file)
Bypass check pass: Audio flows through the Deepgram Voice Agent API (STT/TTS handled by the agent service); DeepgramClient is used correctly via agent.v1.connect() — the proxy only handles the LLM routing layer, not audio
No raw protocol: No WebSocket() or fetch() for Deepgram API contact — the SDK's agent.v1.connect() handles WebSocket

Code quality

✅ Official Deepgram SDK used (deepgram-sdk==6.1.1 — current required version)
✅ tags=["deepgram-examples"] set in Voice Agent settings (demo_agent.py:35)
✅ No hardcoded credentials — all via os.environ
✅ Error handling: missing keys raise RuntimeError, invalid providers return 400, provider failures return 502
✅ Tests import from src/ and test the actual proxy app via TestClient(app)
✅ Server tested with real HTTP requests via FastAPI TestClient
✅ Credential check runs FIRST before SDK imports (lines 1-18 of test file)
✅ No word-list assertions — tests check response structure only (choices, message, usage)

Documentation

✅ README has "What you'll build", env vars table with where-to-get links, install/run instructions
✅ .env.example present with DEEPGRAM_API_KEY and OPENAI_API_KEY
✅ Key parameters table and architecture walkthrough included

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-11

github-actions · 2026-04-11T16:31:23Z

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_health_endpoint PASSED
tests/test_example.py::test_models_endpoint PASSED
tests/test_example.py::test_chat_completions_missing_messages PASSED
tests/test_example.py::test_chat_completions_invalid_provider PASSED
tests/test_example.py::test_chat_completions_openai PASSED
tests/test_example.py::test_provider_header_override PASSED
tests/test_example.py::test_voice_agent_settings_builder PASSED
tests/test_example.py::test_voice_agent_accepts_custom_endpoint_settings PASSED

8 passed, 2 warnings in 2.00s

All 8 tests passed with real credentials. The proxy correctly routes to OpenAI, the Voice Agent SDK connects and receives SettingsApplied from the Deepgram Agent WebSocket, and validation tests cover error paths.

Integration genuineness

✅ Pass — all 6 checks satisfied:

Platform SDK/API imported: OpenAI called via httpx in providers.py; AWS Bedrock via boto3; Deepgram Voice Agent via deepgram-sdk in demo_agent.py
Real API calls made: httpx.post() to api.openai.com, boto3.client.converse() to Bedrock, client.agent.v1.connect() to Deepgram Agent WebSocket
.env.example lists platform credentials: DEEPGRAM_API_KEY and OPENAI_API_KEY both present
Credential guard exits 2: Top of test_example.py exits with code 2 on missing creds before any imports
No bypass: DeepgramClient is used only for the Voice Agent WebSocket (appropriate usage) — audio flows through the Deepgram Agent API, not a raw DeepgramClient transcription call
No raw protocol: No WebSocket(), ws(), or fetch() for audio — SDK's client.agent.v1.connect() used throughout

Code quality

deepgram-sdk==6.1.1 — matches required Python SDK version
tags=["deepgram-examples"] set in AgentV1Settings in demo_agent.py:34
No hardcoded credentials — all secrets from env vars
Error handling: proxy returns 400 for bad provider/missing messages, 502 for provider errors; providers raise on missing keys
Tests import from src/ — proxy.app tested via TestClient, build_settings imported from demo_agent
Server tests spin up FastAPI via TestClient and make real HTTP requests
No word-list assertions — response checks use structure/length only (non-deterministic safe)
Credential check runs FIRST (lines 6-18) before any SDK imports

Documentation

README has "What you'll build" section
All env vars documented with where-to-get links
Install and run instructions present
.env.example present and complete

✓ All checks pass. Ready for merge.

Review by Lead on 2026-04-11

github-actions bot added type:example New example language:python Language: Python integration:voice-agent-multi-provider-proxy Integration: Voice Agent Multi-Provider Proxy labels Apr 8, 2026

github-actions bot mentioned this pull request Apr 8, 2026

build an OpenAI-compatible multi-provider chat completions proxy #195

Closed

feat(examples): add 530 — Multi-Provider Chat Completions Proxy for V…

ba37b26

…oice Agent (Python)

github-actions bot force-pushed the example/530-voice-agent-multi-provider-proxy-python branch from 21dba4e to ba37b26 Compare April 8, 2026 09:01

github-actions bot changed the title ~~[Example] 530 — Voice Agent Multi-Provider Chat Completions Proxy (Python)~~ [Example] 530 — Multi-Provider Chat Completions Proxy for Voice Agent (Python) Apr 8, 2026

github-actions bot added the status:fix-needed Tests failing — fix agent queued label Apr 8, 2026

github-actions bot added status:review-passed Self-review passed and removed status:fix-needed Tests failing — fix agent queued labels Apr 8, 2026

github-actions bot added status:fix-needed Tests failing — fix agent queued and removed status:review-passed Self-review passed labels Apr 10, 2026

fix(examples): use base URL for think.endpoint in 530-voice-agent-mul…

5ce573e

…ti-provider-proxy-python 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>

github-actions bot added status:review-passed Self-review passed and removed status:fix-needed Tests failing — fix agent queued labels Apr 10, 2026

lukeocodes merged commit 7b7ad4a into main Apr 11, 2026

lukeocodes deleted the example/530-voice-agent-multi-provider-proxy-python branch April 11, 2026 18:22

Conversation

github-actions bot commented Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

New example: Multi-Provider Chat Completions Proxy for Voice Agent

What this shows

Required secrets

Tests

Uh oh!

github-actions bot commented Apr 8, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Fixes attempted

Uh oh!

github-actions bot commented Apr 8, 2026

Fix applied

Tests after fix ✅

Uh oh!

github-actions bot commented Apr 9, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 9, 2026

Code Review

Tests ran ✅

Integration genuineness ✅

Code quality ✅

Documentation ✅

Uh oh!

github-actions bot commented Apr 9, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 9, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 9, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 10, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 10, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 10, 2026

Code Review

Tests ran ✅

Integration genuineness

Code quality

Documentation

Uh oh!

github-actions bot commented Apr 10, 2026

Code Review

Tests ran ✅ (7/8) / ❌ (1/8)

Integration genuineness

github-actions bot commented Apr 8, 2026 •

edited

Loading