TTraceAI

The AI Gateway & LLM Observability platform for AI engineers

Access 100+ AI models with a single OpenAI-compatible API key. Observe, debug, and analyze every trace, session, and prompt — with cost, latency, and quality metrics in real time.

Start free — 10k requests/mo
One line to integrate
npx @traceai/sdk init·pip install traceai
traceai.dev/dashboard
Requests+12%
12,481/ 1h
Cost−18%
$24.81today
Avg latency−8%
1.4sp50
ModelEndpointLatencyTokensCost
gpt-4o-mini/chat240ms2,104$0.003
claude-sonnet-4/chat1.2s1,840$0.028
gpt-4o/summarize3.4s3,218$0.048
gemini-2.0-flash/extract180ms612$0.000
claude-haiku-4/classify680ms984$0.002
10kfree requests/mo
100+models supported
1-lineintegration
−38%avg cost savings
One line to integrate

Point your base URL at TraceAI. That's it.

Keep the OpenAI SDK you already use. Swap the baseURL and instantly get logging, routing, fallbacks, and observability across every provider — no SDK rewrite, no data pipeline to maintain.

  • › Works with OpenAI, Anthropic, Gemini, Azure, Bedrock, Ollama & more
  • › JS/TS, Python, cURL, and frameworks like LangChain & Vercel AI SDK
  • › Automatic fallbacks & intelligent routing across providers
quickstart.ts
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://ai-gateway.traceai.dev",
  apiKey: process.env.TRACEAI_API_KEY,
});

const response = await client.chat.completions.create({
  model: "gpt-4o-mini", // claude-sonnet-4, gemini-2.0-flash, or 100+ models
  messages: [{ role: "user", content: "Hello!" }],
});

Everything you need to ship LLM features

From the gateway that routes your requests to the observability that helps you understand them — TraceAI covers the full lifecycle.

100+models, 1 API key

AI Gateway

Access 100+ models through one OpenAI-compatible API with intelligent routing and automatic fallbacks.

0msadded latency

Observe

Inspect and debug traces & sessions for agents, chatbots, and document processing pipelines.

−38%cost, on average

Analyze

Track cost, latency, and quality. Export to PostHog in one line for custom dashboards.

10xfaster iteration

Playground

Rapidly test and iterate on prompts, sessions, and traces directly in the UI.

v1→v8versioned & deployed

Prompt Management

Version prompts using production data. Deploy through the gateway without code changes.

12+spans per session

Sessions & Traces

Group related requests into user sessions to trace multi-step conversations end-to-end.

320cases from real traffic

Datasets & Fine-tuning

Build datasets from request logs for evaluation, fine-tuning, or batch analysis.

SOC 2& GDPR compliant

Enterprise Ready

Self-host on your own cloud with Docker or Helm. SSO, audit logs, and RBAC included.

Agent tracing

Find the one span that cost you 18 seconds

Multi-step agents as waterfall trees. Critical path, cost attribution, and latency outliers — highlighted automatically.

  • Critical path — 78% of wall-clock time in a single span, flagged automatically
  • Cost attribution — per LLM call, per tool, per retrieval step
  • Retry & error spans as first-class citizens, not hidden in logs
trace_8812 · support agent
8.24s critical
└
agent.run
8.24s
└
classify_intent
520ms
└
kb_search
680ms
└
summarize_tickets · v7
5.8s
└
llm.sonnet-4
5.4s
└
format_reply
480ms
critical path highlighted·cost: $0.083·5 spans

Works with the tools you already use

100+ inference providers and every major framework — through a single, OpenAI-compatible API.

Inference providers

OpenAIAnthropicGeminiAzure OpenAIAWS BedrockVertex AIOllamaMistralCohereTogetherGroqDeepSeek+90 more

SDKs & frameworks

TypeScript SDKPython SDKLangChainVercel AI SDKOpenLLMetrycURL

How it works

Three steps from your first request to full observability.

01

Route through the gateway

Send requests to the TraceAI AI Gateway instead of your provider directly. One API key unlocks 100+ models with intelligent routing and automatic fallbacks.

02

We log everything

The proxy captures request and response data asynchronously via a queue — no latency added to your users. Metadata goes to Postgres, analytics to ClickHouse, large payloads to object storage.

03

Observe, analyze, iterate

Inspect traces and sessions, track cost & latency, version prompts, build datasets, and test in the playground — all from one dashboard.

Simple, transparent pricing

Start free with 10k requests a month. Upgrade when you scale. Self-host anytime at no cost.

Free

For getting started and small projects.

$0/ forever
Start free
  • 10,000 requests / month
  • Dashboard, requests & sessions
  • Prompts, datasets & playground (10 runs)
  • 1 alert
  • Community support

Pro

For teams shipping LLM features to production.

Custom/ scaled to usage
Start Pro trial
  • Unlimited requests
  • Unlimited playground, alerts & experiments
  • Evaluators & fine-tuning
  • Key vault & customer portal
  • Priority support

Enterprise

For organizations with scale & compliance needs.

Talk to us
Contact sales
  • On-prem / VPC deployment (Helm)
  • SOC 2 & GDPR compliance
  • Custom ETL & SSO
  • Dedicated support & SLA

Reasonable questions

Everything you need to know before integrating.

Swap your provider SDK's baseURL to point at the TraceAI gateway. Same surface, same types. We record the full request and response on the wire, with no extra round-trip and no sampling by default. Works with the OpenAI SDK, Anthropic SDK, LangChain, Vercel AI SDK, and more.

Ship your first traced request in 2 minutes

No credit card. No infrastructure to manage. Just point your base URL at TraceAI and start observing.

Get started free