
Connection map / 20 adapters
Providers, frameworks, runtimes and alert channels all converge on one OpenTelemetry ingest surface.
Adapters
Ingest spans through the standard OTLP HTTP JSON protocol. Any OTel-instrumented library or exporter works out of the box.
Tracing
Drop-in LLM observability instrumentation for OpenAI, Anthropic, Cohere, and more.
Tracing
Trace LangChain agents, chains, and tool calls through the built-in OTel callback handler.
Frameworks
Monitor CrewAI multi-agent workflows through Python SDK instrumentation.
Frameworks
Trace LlamaIndex queries, retrieval and agent steps through Python SDK instrumentation.
Frameworks
Trace Mastra agents and workflows through the TypeScript SDK or OpenTelemetry.
Frameworks
Trace Pydantic AI agent runs and tool calls through the Python SDK or OpenTelemetry.
Frameworks
Trace agent runs, handoffs and tool calls from the OpenAI Agents SDK.
Frameworks
Trace AutoGen conversations and tool execution through OTel instrumentation or the Python decorator.
Frameworks
Capture model, latency, tokens, and cost for every OpenAI API call automatically.
Providers
Trace Claude model calls, tool use, streaming, and extended-thinking spans.
Providers
Trace streaming calls, tool use, and multi-step generations from the Vercel AI SDK.
Providers
Server-side tracing for route handlers, Server Actions, and middleware through the TypeScript SDK.
Runtimes
Wrap FastAPI, Flask, or ordinary Python functions for automatic span collection.
Runtimes
Track cache hits, misses, and rate-limiter state as spans in the agent pipeline.
Infrastructure
Route alerts and failed-trace notifications into Slack channels through webhooks.
Alerts
Route alerts and failed-trace notifications into a Teams channel through an incoming webhook or Power Automate flow.
Alerts
Escalate high-severity agent failures into the on-call workflow.
Alerts
Send spans, alerts, cost thresholds, and orchestration events to any HTTP endpoint.
Alerts
Give AI agents and MCP-compatible clients read access to Tracify's documentation as structured resources.
Agent tooling
Setup guides
Three steps
Set the OTLP endpoint and your project key. Nothing else is required.
Use any OTel-instrumented library, wrap your agent with the TypeScript or Python SDK, or post spans over plain HTTP.
Spans arrive as one trace per run, with cost, tokens and failures attached.
import { traceAgent, span } from "tracify-sdk";
export const answer = traceAgent(async (q: string) => {
const docs = await span("docs.search", () => search(q), { type: "tool" });
return span("chat", () => callModel(q, docs), { type: "llm" });
});Questions
Anything that speaks OpenTelemetry works through the OTLP endpoint. LangChain, CrewAI, LlamaIndex, AutoGen, Mastra, Pydantic AI and the OpenAI Agents SDK are covered through the Python and TypeScript SDKs or OTel instrumentation.
Usually a few lines: install the SDK, set a project key and wrap the client or add the OTel exporter. Provider calls such as OpenAI and Anthropic record model, latency, tokens and cost automatically.
Yes. Point an OTLP HTTP JSON exporter at Tracify with your project key. See the ingest reference in the docs for the endpoint and headers.
Yes. Project alerts can post failures and threshold breaches to a validated Slack webhook.
If it can export OpenTelemetry spans it already works. Otherwise use the contact page to request an adapter.
If it speaks OpenTelemetry it already works. If it does not, tell us and we will look.
