beacon-sdk 0.1.0

Instrumentation SDK for Beacon — Chrome DevTools for AI Agents

beacon-sdk

Instrumentation SDK for Beacon.

Add tracing to Python agents with minimal changes and inspect traces in Beacon UI.

Quickstart

pip install beacon-sdk[openai]

import beacon_sdk
from openai import OpenAI

beacon_sdk.init()
client = OpenAI()

@beacon_sdk.observe
def run_agent(question: str) -> str:
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": question}],
    )
    return response.choices[0].message.content or ""

run_agent("What is the capital of France?")

Start Beacon (make dev) and open http://localhost:5173.

Integrations

Integration	Behavior
OpenAI	auto-patched chat completions (sync/async + streaming + tool calls)
Anthropic	auto-patched messages (sync/async + streaming + tool use)
Google Gemini	auto-patched generate_content (sync/async + streaming)
CrewAI	auto-patched Crew.kickoff() with callback injection
AutoGen	auto-patched ConversableAgent.generate_reply() and GroupChat.run()
LlamaIndex	auto-patched BaseQueryEngine.query() and BaseRetriever.retrieve()
Ollama	auto-patched chat() and generate() (native client)
LiveKit Agents	auto-patched AgentSession voice lifecycle + key session events
Playwright	auto-patched page actions
subprocess	auto-patched run and check_output
LangChain	callback handler (BeaconCallbackHandler)

File operation patching (builtins.open) is opt-in via BEACON_PATCH_FILE_OPS=true.

Install Extras

pip install beacon-sdk
pip install beacon-sdk[openai]
pip install beacon-sdk[anthropic]
pip install beacon-sdk[playwright]
pip install beacon-sdk[livekit]
pip install beacon-sdk[all]

Configuration

Env var	Default	Meaning
BEACON_BACKEND_URL	http://localhost:7474	backend ingestion URL
BEACON_ENABLED	true	enable/disable tracing
BEACON_AUTO_PATCH	true	enable/disable auto-patching
BEACON_LOG_LEVEL	WARNING	SDK logging level
BEACON_PATCH_FILE_OPS	false	enable file operation patch

init() options:

beacon_sdk.init(
    backend_url="http://localhost:7474",
    auto_patch=True,
    enabled=True,
    exporter="auto",  # auto | async | sync
)

Public API

• init()
• observe
• get_current_span()
• get_tracer()
• flush()
• shutdown()
• Span, SpanType, SpanStatus, BeaconTracer

OTLP Ingestion (SDK-Free)

Beacon accepts traces in standard OpenTelemetry Protocol (OTLP) JSON format. This lets you send traces from any OTEL-instrumented application without installing the Beacon SDK.

Endpoint: POST /v1/otlp/traces

Example payload:

{
  "resourceSpans": [
    {
      "resource": {
        "attributes": [
          { "key": "service.name", "value": { "stringValue": "my-agent" } }
        ]
      },
      "scopeSpans": [
        {
          "scope": { "name": "my-instrumentation", "version": "1.0.0" },
          "spans": [
            {
              "traceId": "abc123",
              "spanId": "span001",
              "name": "llm_call",
              "kind": 1,
              "startTimeUnixNano": "1700000000000000000",
              "endTimeUnixNano": "1700000001500000000",
              "attributes": [
                { "key": "span_type", "value": { "stringValue": "llm_call" } },
                { "key": "llm.model", "value": { "stringValue": "gpt-4o" } },
                { "key": "llm.tokens.total", "value": { "intValue": "1500" } },
                { "key": "llm.cost_usd", "value": { "doubleValue": 0.023 } }
              ],
              "status": { "code": 1 }
            }
          ]
        }
      ]
    }
  ]
}

curl example:

curl -X POST http://localhost:7474/v1/otlp/traces \
  -H "Content-Type: application/json" \
  -d '{
    "resourceSpans": [{
      "scopeSpans": [{
        "spans": [{
          "traceId": "test-trace-001",
          "spanId": "test-span-001",
          "name": "my-llm-call",
          "startTimeUnixNano": "1700000000000000000",
          "endTimeUnixNano": "1700000001000000000",
          "attributes": [
            {"key": "span_type", "value": {"stringValue": "llm_call"}},
            {"key": "llm.model", "value": {"stringValue": "gpt-4o"}}
          ],
          "status": {"code": 1}
        }]
      }]
    }]
  }'

Response: { "accepted": 1, "rejected": 0 }

Attribute value types: stringValue, intValue, doubleValue, boolValue, arrayValue

Status codes: 0 = Unset, 1 = OK, 2 = Error

Valid span_type values: llm_call, tool_use, agent_step, shell_command, browser_action, custom (default)