Product

Six things AgentOps Monitor does for your agents.

Trace, cost, cap, audit, ingest, and overview. Every feature exists to answer one question: what did the agent do, and what did it cost.

Start monitoring See pricing

Trace every tool call and LLM decision

Every run becomes a replayable timeline — model output, tool calls, latencies, errors. Click any span to see the prompt, the model output, and what the agent changed in the world.

+0 ms      refund-triage-v3 start +12 ms     llm openai.chat        8,204 tok +612 ms    tool vectorstore.query +820 ms    tool stripe.refunds.create  HELD +5,402 ms  tool retries × 3 +11,008 ms llm openai.chat        64,118 tok +18,402 ms tool slack.postMessage +48,219 ms end · budget exceeded

Track every dollar an agent spends

Tokens in and out, USD per span, total per run. The dashboard surfaces the runs that cost the most so on-call sees the bill before the invoice arrives.

Cap the spend with hard budget limits

Set caps at the workflow, the agent, or the user. A run stops at the limit you set — not at the limit your CFO finds out about.

refund.cap_per_run_usd: 50 require_approval_above: 250 stripe.refunds.create: hold_above_usd: 100 loop.max_retries: 3 halt_after_seconds: 60

Audit every decision

The audit trail captures prompt hashes, model output, tool calls, human approvals, and policy decisions. SOC 2 / HIPAA / GDPR badges are not claimed on this page. The audit trail is the product.

Ingest from any framework

The SDK ships ingestion for the frameworks you already run. Planned adapters: LangChain, CrewAI, OpenAI SDK, Anthropic.

See integrations

See your fleet at a glance

Total runs, active keys, recent runs with spans, cost, and status. One dashboard for the whole agent fleet.

Start free