Skip to content

Installation & Quick Start

Jump to bottom
Niccanor Dhas edited this page Feb 22, 2026 · 1 revision

Installation & Quick Start

Get your first LLM trace flowing in under 5 minutes.

Prerequisites

  • Python 3.8+
  • A running tmam instance — either [self-hosted](Self-Hosting-with-Docker) or the [cloud](https://cloud.tmam.ai)
  • Your tmam public key and secret key (from Settings → API Keys in the dashboard)

Step 1 — Install the SDK

pip install tmam

Step 2 — Initialize tmam

Call init() once at application startup, before any LLM calls:

from tmam import init

init(
    url="http://localhost:5050/api/sdk",  # your tmam server endpoint
    public_key="pk-tmam-xxxxxxxx",
    secrect_key="sk-tmam-xxxxxxxx",
    application_name="my-app",
    environment="production",
)

Cloud users: replace the url with https://cloud.tmam.ai/api/sdk and use your cloud API keys.

Step 3 — Use your LLM normally

No code changes required. tmam wraps supported libraries automatically:

from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What is observability?"}]
)
print(response.choices[0].message.content)

Every call is now automatically traced with token counts, cost, latency, model name, and optionally message content.

Step 4 — View in the dashboard

Open the dashboard at http://localhost:3001 and navigate to Observations → Requests to see your first traces.

Complete init() options

from tmam import init

init(
    # Required
    url="http://localhost:5050/api/sdk",
    public_key="pk-tmam-xxxxxxxx",
    secrect_key="sk-tmam-xxxxxxxx",

    # Identify your app
    application_name="my-app",       # default: "default"
    environment="production",         # default: "default"

    # Features
    capture_message_content=True,    # trace prompt/completion text (default: True)
    collect_gpu_stats=False,         # enable GPU metrics (default: False)
    disable_metrics=False,           # disable OTel metrics (default: False)
    disable_batch=False,             # send spans immediately (default: False)

    # Selective instrumentation
    disabled_instrumentors=["openai"],  # skip specific providers

    # Guardrails
    guardrail_id="your-guardrail-id",  # default guardrail for Detect calls

    # Advanced
    pricing_json="/path/to/pricing.json",  # custom model pricing
)

See the full [SDK Initialization](SDK-Initialization) page for detailed option descriptions.

What gets traced automatically?

Once initialized, tmam auto-detects installed libraries and instruments them. No extra imports needed.

✅ OpenAI detected          → OpenAI calls instrumented
✅ anthropic detected       → Anthropic calls instrumented
✅ langchain detected       → LangChain chains instrumented
✅ chromadb detected        → Chroma queries instrumented

See [Supported Integrations](Supported-Integrations) for the full list.

Next Steps

Clone this wiki locally