Real-time traces, automatic kill-switch, and loop detection — in 2 lines of code.
Free up to 1,000 traces/mo · No credit card · Instrument in 2 lines
I need to check the user's subscription status before processing the refund request.
get_subscription(user_id: "usr_4829")
{"status": "active", "plan": "pro", "eligible_for_refund": true}
User is eligible. Processing refund now
Works with any agent — LangChain, CrewAI, raw OpenAI calls, or your own loop.
from traceagently import TraceAgently
ta = TraceAgently(api_key="ta_live_...")
with ta.trace(agent_id="support-bot",
task="Refund order #123") as t:
t.thought("Checking order status")
t.tool_call("get_order", {"id": 123})
t.tool_result({"status": "delivered"})
t.final_response("Order was delivered")import { TraceAgently } from 'traceagently'
const ta = new TraceAgently({ apiKey: 'ta_live_...' })
const trace = await ta.startTrace({
agentId: 'support-bot',
task: 'Refund order #123'
})
await trace.thought('Checking order status')
await trace.toolCall('get_order', { id: 123 })
await trace.end('Order was delivered')The only observability platform that intervenes — not just observes.
Set a per-trace cost limit. The moment a runaway agent exceeds it, it's automatically killed — no more API calls, no more bill. Never wake up to a $1,000 charge again.
Catches infinite loops before they spiral. If your agent calls the same tool N times in a row, it's killed immediately. Configurable threshold — default is 5 consecutive calls.
Watch every thought, tool call, and response stream live as your agent runs. Full event timeline with timestamps, token counts, and cost per step.
Side-by-side diff of any two runs. See exactly what changed in cost, tokens, duration, and event count. Mark a good run as Golden and compare all future runs against it.
One click sends the full trace to Claude, which reads every step, diagnoses the failure, and streams back an exact fix. No copy-pasting logs into ChatGPT.
Token usage and estimated cost calculated per trace automatically. Know exactly what each agent run costs before it becomes a problem.
Get notified the moment a kill-switch fires or a loop is detected. Uses your account email — zero configuration, no webhooks to set up.
Recurring errors across all your runs are automatically grouped and ranked. Spot systemic failures before they reach your users.
Works with LangChain, CrewAI, raw OpenAI, Claude, Gemini, or any custom agent loop. 2-line SDK integration in Python or TypeScript.
Start free. One plan when you're ready for more.
No credit card required
Limits
Features
$0.50 per 1k traces over limit
Limits
Features
From zero to live traces in under 2 minutes.
Install the SDK
pip install traceagently # Python
npm install traceagently # TypeScriptGet your API key from the dashboard
ta_live_••••••••••••••••••••••••Wrap your agent
from traceagently import TraceAgently
ta = TraceAgently(api_key="ta_live_...")
with ta.trace(
agent_id="my-agent", task=user_task
) as t:
t.thought(llm_response)
t.tool_call(tool_name, args)
t.final_response(result)TraceAgently is a real-time observability tool for AI agents. It captures every thought, tool call, tool result, and error your agent makes and shows them live in a dashboard. Works with OpenAI, Claude, Gemini, LangChain, CrewAI, or any custom agent loop you wrote yourself.
Regular logs dump flat text. They have no understanding of agent loop structure. TraceAgently captures each event type in sequence so you see exactly what your agent was thinking at each step and why it failed. It is built specifically for agents, not web servers.
TraceAgently works with OpenAI, Anthropic Claude, Google Gemini, LangChain, LangGraph, CrewAI, AutoGen, and any custom Python or TypeScript agent loop. It is completely framework-agnostic. If you can add a function call to your agent, you can use TraceAgently.
Yes. The free tier gives you 1,000 traces per month, 7-day data retention, and 1 agent. No credit card required. When you're ready for more, the Pro plan is $49/month for 1 million traces, 90-day retention, unlimited agents, kill-switch, loop detection, email alerts, trace comparison, and Magic Fix AI.
A trace is one agent run — the unit developers think in. Events are internal implementation details that vary wildly between agents: a simple bot might produce 3, a complex one 300. Billing by traces means the price is predictable. Per-seat billing punishes growth. We give you 1 million traces for $49 flat — LangSmith charges $39 per seat for far fewer runs.
Run pip install traceagently for Python or npm install traceagently for TypeScript. Sign up to get your API key, then add 2 lines to your agent code. Your first trace appears in the dashboard within seconds.
Yes. Your trace data is scoped to your account and never visible to other users. We do not sell your data or use it to train models. Data is automatically deleted after your retention window (7 days on Free, 90 days on Pro).