Migrating from Datadog LLM Observability

This page is the mechanical walkthrough for moving AI-agent tracing off Datadog LLM Observability: swap the ddtrace.llmobs init for Tracelane’s, replace the proprietary decorators with OpenTelemetry GenAI semconv spans, and rebuild the handful of views you relied on. You can also run both side by side — see Keeping Datadog for infrastructure APM. We do not restate Datadog’s feature set or pricing here — both move, and their own documentation is the only accurate source for them. Below is what Tracelane brings, so you can compare it against your current bill and plan yourself.

What Tracelane gives you

Capability	Tracelane
LLM call tracing	OTel GenAI semconv spans — portable to any OTel-compatible backend
Agent-specific signals	Inline heuristic rails at the gateway; MCP/agent-tool detectors fire on tool traffic, not chat completions
Inline guardrails	Rails run in-request and verdicts are written to the ledger. The predictive layer is observe-first by default — a `Block` is recorded and the request still proceeds unless enforcement is explicitly enabled
Prompt versioning + eval gates	Prompt promotion live; rollback is EWMA drift on metrics you post, not autonomous; eval gating live — a promotion without a passing run returns `409` unless you record an attributed override
Time-travel replay	Read-only re-render; cross-model re-execution is on the roadmap
BYOK gateway	205 providers, 0 % markup on your provider keys
Tamper-evident audit ledger	Hash chain + Merkle + Ed25519 per-tenant signing (live); resolved Sigstore Rekor v2 inclusion proof on anchored batches (best-effort); universal per-run anchoring on the roadmap
Self-hostable	Apache 2.0, single-node Docker Compose (`infra/self-host/docker-compose.yml`). No Helm chart is published
Pricing model	Six usage meters, including ingest bytes — see Pricing

Step 1 — Install Tracelane

pip install tracelane        # Python
npm install @tracelanedev/sdk   # TypeScript

Step 2 — Remove Datadog LLM Observability init

Before:

from ddtrace.llmobs import LLMObs

LLMObs.enable(
    ml_app="my-agent",
    api_key=os.environ["DD_API_KEY"],
)

After:

import tracelane

tracelane.init(
    endpoint="https://gateway.tracelane.dev",  # or an OTLP receiver you run
    api_key="tlane_...",                       # needs the `ingest` scope
)
tracelane.auto_instrument()  # prints attachment outcomes for supported integrations

auto_instrument() attempts Anthropic, OpenAI, LiteLLM, LangGraph, and Claude Code. The langgraph extra enables graph and child capture through LangChain callbacks. For other adapters, call the explicit instrument_* function after constructing the object (see the Python SDK). On Tracelane Cloud the OTLP endpoint is the gateway itself and the key needs the ingest scope; self-hosting, it is the receiver you run. Remove DD_API_KEY, DD_LLMOBS_ENABLED, DD_LLMOBS_ML_APP from your environment.

Step 3 — Replace Datadog spans with OTel spans

Datadog uses a proprietary span API. Tracelane uses OpenTelemetry GenAI semconv — portable across any OTel-compatible backend.

# Datadog
from ddtrace.llmobs.decorators import llm

@llm(model_provider="anthropic", model_name="claude-sonnet-4-6", name="my-call")
def invoke(prompt):
    ...

# Tracelane — standard OpenTelemetry API (GenAI semconv attribute keys)
from opentelemetry import trace

tracer = trace.get_tracer("my-agent")
with tracer.start_as_current_span("my-call") as s:
    s.set_attribute("gen_ai.system", "anthropic")
    s.set_attribute("gen_ai.request.model", "claude-sonnet-4-6")
    result = client.messages.create(...)

Or wrap the client with its instrument_* function — no manual span code needed:

import tracelane
from tracelane import instrument_anthropic

tracelane.init(endpoint="http://localhost:4318", api_key="tlane_...")

from anthropic import Anthropic
client = Anthropic()
instrument_anthropic(client)  # now this client's calls are traced

Step 4 — Migrate dashboards

Datadog dashboards use proprietary widgets. Tracelane provides:

Traces view — transcript-spine span tree for deep agent traces
SLO dashboard — error rate, latency, token spend per model/provider. The availability target is the product default (99.9 %); a per-tenant SLO target is not configurable yet
Eval suite — 20 conformance evals, run in CI (no in-app scoreboard yet)
Audit ledger — in-app ledger view and NDJSON export (/audit)

If you have custom Datadog monitors, the closest equivalent is an alert rule: a threshold on one of six metrics — error_rate, burn_rate, latency_p95, overhead_p99, cost_usd, quota_pct — delivered to your own Slack- or Discord-compatible webhook. Alert rules are entitlement-gated; check whether your plan includes them. Anything outside those six is an eval assertion or nothing.

Keeping Datadog for infrastructure APM

Tracelane traces AI-specific spans (LLM calls, tool calls, A2A handoffs). You can run both simultaneously — Tracelane handles the AI layer, Datadog handles the infrastructure layer. They are not mutually exclusive.

# Both active at the same time
import ddtrace; ddtrace.patch_all()   # infra APM
import tracelane
tracelane.init(endpoint="http://localhost:4318", api_key="tlane_...")  # AI agent layer
tracelane.auto_instrument()

Competitor details verified as of 2026-08-26 against public documentation and public repositories. Where a capability is marked absent, we found none documented at that date — not that the vendor lacks it. If anything here is inaccurate, please open an issue.

Get started

SDKs

Integrations

Concepts

Features

Tools

Audit & compliance

Operations

Migrations

Comparisons

Migrating from Datadog LLM Observability

What Tracelane gives you

Step 1 — Install Tracelane

Step 2 — Remove Datadog LLM Observability init

Step 3 — Replace Datadog spans with OTel spans

Step 4 — Migrate dashboards

Keeping Datadog for infrastructure APM

​What Tracelane gives you

​Step 1 — Install Tracelane

​Step 2 — Remove Datadog LLM Observability init

​Step 3 — Replace Datadog spans with OTel spans

​Step 4 — Migrate dashboards

​Keeping Datadog for infrastructure APM

​Related

What Tracelane gives you

Step 1 — Install Tracelane

Step 2 — Remove Datadog LLM Observability init

Step 3 — Replace Datadog spans with OTel spans

Step 4 — Migrate dashboards

Keeping Datadog for infrastructure APM

Related