emotion-machine 0.0.7

Infra for AI Companions

Emotion Machine Python SDK

Official Python helper for the Emotion Machine Companion API. It wraps the /v1 endpoints documented in docs/client-companion-api-plan.md so you can provision companions, ingest knowledge, and chat/stream with them in just a few lines of code.

Installation

pip install emotion-machine

The client depends on httpx and targets Python 3.9+.

Quickstart

1. Export your project API key (project-scoped):

export EM_API_KEY="emk_prod_...."
export EM_API_BASE_URL="https://api.emotionmachine.ai"  # or http://localhost:8100 for local dev

2. Bootstrap a companion, ingest curated knowledge, and chat:

from emotion_machine import EmotionMachine

client = EmotionMachine()  # reads EM_API_KEY / EM_API_BASE_URL

# Create a fresh companion
companion = client.create_companion(
    name="Luteal Support Coach",
    description="Helps users track luteal phase cravings",
    config={
        "system_prompt": {
            "full_system_prompt": "You are an encouraging health coach."
        }
    },
)
companion_id = companion["id"]

# Optionally shape the profile schema for per-user traits
client.upsert_profile_schema(
    companion_id,
    schema={
        "type": "object",
        "properties": {
            "craving_intensity": {"type": "integer", "minimum": 0, "maximum":5}
        },
    },
)

# Ingest curated luteal-phase knowledge via the built-in key
job = client.ingest_knowledge(
    companion_id,
    payload_type="json",
    key="data_id_x",
)
job_result = client.wait_for_job(job["id"], timeout=20)
assert job_result["status"] == "succeeded", job_result

# Or upload & ingest your own JSON/Markdown/TXT file in one step.
# The helper uploads the file, kicks off ingestion, waits for completion,
# and raises if the job fails.
result = client.ingest_file(
    companion_id,
    file_path="important_app_related_knowledge.jsonl",
    payload_type="json",
)
print(result["job"]["status"])  # -> "succeeded"

# Run a synchronous chat completion
completion = client.chat_completion(
    companion_id,
    message="Hi! I'm feeling intense salt cravings today, what should I know?",
       external_user_id="user-123",
   )

print(completion["choices"][0]["message"]["content"])

# Stream responses (Server-Sent Events) and collect message chunks
stream = client.chat_stream(
    companion_id,
    message="Can you summarise key luteal phase symptoms?",
    external_user_id="user-123",
)
for event in stream:
    if event["event"] == "delta":
        chunk = event["data"]["choices"][0]["delta"].get("content", "")
        if chunk:
            print(chunk, end="", flush=True)
    elif event["event"] == "done":
         conversation_id = event["data"]["conversation_id"]

# Retrieve the full conversation transcript
transcript = client.get_conversation(conversation_id)
for message in transcript["messages"]:
    print(f"{message['role']}: {message['content']}")

# Upload an image and ask about it
image_result = client.upload_image(
    companion_id,
    conversation_id,
    file_path="food_photo.jpg",
)
image_id = image_result["image_id"]
print(f"Image description: {image_result['description']}")

# Send a message referencing the image
response = client.chat_completion(
    companion_id,
    message="What do you think about this meal?",
    external_user_id="user-123",
    conversation_id=conversation_id,
    image_ids=[image_id],
)
print(response["choices"][0]["message"]["content"])

# List the most recent conversations for a specific tester/user cohort
recent_sessions = client.list_conversations(
    companion_id,
    limit=25,
    external_user_id="user-123",
)
print(f"Found {len(recent_sessions)} saved sessions for user-123")

# Filter options include `external_user_prefix="beta-"` for cohort-level filtering.

# Create a voice session for real-time audio
session = client.create_session(
    companion_id,
    voice_config={
        "pipeline_type": "openai-realtime",
        "voice_name": "alloy",
        "temperature": 0.7,
        "realtimeModel": "gpt-realtime-mini-2025-10-06",
    },
)
print(f"Connect to WebSocket: {session['ws_url']}")
# Use a WebSocket library (e.g., websockets) to stream PCM audio

3. Tidy up when finished:

client.close()

or use with EmotionMachine() as client: to auto-close the HTTP session.

Knowledge management tips

• client.list_knowledge_assets(companion_id) shows the latest uploads plus their statuses (ready, superseded, etc.).
• Uploading a file with the same filename automatically replaces older ingestions for that companion’s vector store—no manual delete required.
• client.search_knowledge(..., mode="semantic" | "keyword" | "hybrid") lets you compare retrieval strategies against the same dataset.
• client.wait_for_job() and client.ingest_file() raise KnowledgeJobFailed if OpenAI reports the ingestion job as failed, so you can catch mistakes early in CI.

API Coverage

Resource	Method	SDK helper
Companions	GET /v1/companions	client.list_companions()
	POST /v1/companions	client.create_companion(...)
	GET /v1/companions/{id}	client.get_companion(id)
	PATCH /v1/companions/{id}	client.update_companion(...)
	DELETE /v1/companions/{id}	client.delete_companion(id)
Profile Schema	PUT /v1/companions/{id}/profile-schema	client.upsert_profile_schema(...)
	GET /v1/companions/{id}/profile-schema	client.get_profile_schema(...)
Knowledge	POST /v1/companions/{id}/knowledge	client.ingest_knowledge(...)
	GET /v1/knowledge-jobs/{job_id}	client.knowledge.get_job(job_id)
Chat	POST /v1/companions/{id}/chat	client.chat_completion(...)
Chat (stream)	POST /v1/companions/{id}/chat/stream	client.chat_stream(...)
Conversations	GET /v1/companions/{id}/conversations	client.list_conversations(...)
Conversations	GET /v1/conversations/{conversation_id}	client.get_conversation(...)
Images	POST /v1/companions/{id}/conversations/{id}/images	client.upload_image(...)
	GET /v1/companions/{id}/conversations/{id}/images	client.list_images(...)
Voice Sessions	POST /v1/sessions	client.create_session(...)
	PATCH /v1/sessions/{id}	client.update_session(...)

All helpers raise emotion_machine.APIError on non-success HTTP codes. Inspect e.status_code and e.payload for diagnostics.

Development

cd packages/pip-emotion-machine
pip install -e .[dev]

The package ships from src/emotion_machine. Update pyproject.toml to bump versions.