Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.tracelm.ai/llms.txt

Use this file to discover all available pages before exploring further.

Official Python SDK for reliability-first observability in production agents.

​
Install

pip install tracelm

​
Quick Start

from tracelm import TraceLM

tracelm = TraceLM(
    api_key="lt_your_tracelm_key",
    openai_api_key="sk-your-openai-key",
)

response = tracelm.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
)

​
Providers

TraceLM supports provider="openai" | "anthropic" | "google" and can route through /v1/chat/completions with provider-normalized behavior. You can also use provider-specific clients:
  • AnthropicTraceLM
  • GeminiTraceLM

​
Reliability-First Onboarding

  1. Create project + API key.
  2. Send one traced request.
  3. Open /dashboard/first-run.
  4. Confirm your first reliability signal.

​
Task Grouping + Detection

with tracelm.task(name="booking_flow", user_id="user_123") as task:
    tracelm.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "Book a flight"}],
    )

    result = task.complete()  # run_detection=True by default

    if result:
        print(result.loops.detected)
        print(result.failures.explicit)
        print(result.context.total)
        print(result.tool_failure_count)
        print(result.context_failure_count)

​
Manual Detection (No Completion)

detection = task.detect()
print(detection.loops.patterns)
print(detection.failures.semantic)
print(detection.context.lost_context)

​
Conversation Tracking

with tracelm.conversation(user_id="user_123", session_id="sess_1") as conv:
    with conv.task(name="turn_1") as task:
        tracelm.chat.completions.create(model="gpt-4o-mini", messages=[{"role": "user", "content": "Hi"}])
        task.complete()

​
Query APIs

tasks = tracelm.list_tasks(has_issues=True)
item = tracelm.get_task("task-uuid")
convs = tracelm.list_conversations(user_id="user_123")
failures = tracelm.get_conversation_context_failures("conv-uuid")
Use these queries to move from individual traces to execution-level diagnosis.

​
Environment Variables

TRACELM_API_KEY=lt_...
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...
TRACELM_BASE_URL=https://api.tracelm.ai

​
Notes

  • task.complete() returns None if the backend task record does not exist yet.
  • Use run_detection=False for completion without re-running detection.
  • Detection results are stored in task metadata under detection_results.