AgentOps MonitorStart free

Product

Six things AgentOps Monitor does for your agents.

Trace, cost, spans, 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

The example run beside this text works like a run in your dashboard: select any span to open it.

Example · 02:14:11Z · run r_9f2c1a4e · refund-triage-v3Run cost $0.190
Event
6 of 8
Type
llm
Model / tool
openai / gpt-4o
Tokens
64,118 tok
Duration
7,394 ms
Cost
$0.164
Status
ok

Found it. After three retries the agent sent a 64,118-token context back to the model. This one call is $0.164 of the run's $0.190 (86%). At 2,400 runs a night, that is about $394 by morning.

Trace your own agent

Eight timeline events are six grouped span rows plus run start/end markers. The six rows represent eight recorded spans because three Stripe retry attempts share one row. HELD is a status reported by your agent at ingest; AgentOps Monitor records it and does not hold the call.

Works today

What you get on every plan.

Works today

Trace every tool call and LLM decision

Every run becomes a time-ordered list of spans: type, tool or model, status, duration, and cost. Open any span to see the input and output your agent sent with it.

Works today

Track every dollar an agent spends

Tokens in and out and USD cost for every run, and the USD cost of every span, as your agent reports them. The run list shows the cost of every run. A runaway call is recorded here. It is not stopped, and spend caps and cost alerts are not built yet.

Works today

Every span, with the cost attached

Each span records its model, provider, tool name, status, duration, USD cost, and the raw input and output your agent reported, so an expensive loop is visible before it reaches an invoice.

Works today

Audit what actually ran

The trace keeps the model and provider behind each call, which tool was invoked, the input and output payloads, and the cost and duration of every span. Human approval workflows and policy decisions are not captured yet - they are on the roadmap below. SOC 2 / HIPAA / GDPR badges are not claimed on this page.

Works today

Ingest from any framework

Send traces from any framework with a plain HTTP POST to /api/ingest. First-party adapters are not shipped yet.

See integrations →

Works today

See your fleet at a glance

Total runs, API keys, and your most recent runs with status, tokens, and cost. Open a run to see its spans.

What one request stores

One POST per run. Everything above is built from these fields.

POST /api/ingest
  run   → agent_name, status, tokens_in/out, cost_usd
  span  → model, tool_name, status, duration_ms, cost_usd, input, output

Start free Read the docs

Planned

Planned / Roadmap

These capabilities are planned and are not built yet. They are listed here so the roadmap is public, not implied.