Prove what your AI agents did.

Which agents run, who owns them, what they called and what they cost. On the same platform that watches the servers they run on.

Unknown is never zero.

The four questions every audit asks

  • Which agents run?

    The register lists every agent, claimed or not.

    See it
  • Who owns them?

    Each agent carries an owner, a purpose and a risk level.

    See it
  • What did they do?

    Every run is kept as its own record.

    See it
  • Can you prove it?

    Every finding shows the query behind it.

    See it

REGISTER

Every agent has an owner.

Claim each agent with an owner, a purpose, a risk level and an environment. Agents running in your name that nobody has claimed are listed on their own, so shadow AI shows up instead of hiding.

The agent register, with unclaimed agents listed on their own

RUNS

Every run, on the record.

Spans, model calls, tool calls and cost for each run. Runs are kept as their own records, so they outlive the traces they came from.

One agent run, with its spans, model calls, tool calls and cost

EVIDENCE

Every finding shows its query.

When investigations are switched on for your workspace, they gather evidence from your metrics, logs and traces. Each item shows the query it ran. Re-run it, and shoofi tells you whether it still matches the evidence or the data has changed since.

One investigation as a timeline: where it came from, the window it looked at, its ten queries, and a hypothesis marked not verified

ALERTING

Judged against its own rhythm.

shoofi learns how often each agent reports and alerts when it goes quiet for far longer than usual. A nightly job and a five-minute job are held to different bars. When shoofi cannot see your agents at all, that is its own alert, never a green light.

Nightly jobunknownFive-minute jobexpected window

FOUNDATION

And it watches everything underneath.

Hosts, containers, Kubernetes workloads, logs, traces, URL uptime and TLS certificates, with a maintained catalogue of alert rules you switch on and tune. Alerts arrive by email or Telegram.

  1. 1Pick what to watch.
  2. 2Copy the collector config we generate for you.
  3. 3See the first data arrive.
The hosts view, with the machines this workspace watches

HONESTY

The list most vendors leave out

  • It proposes hypotheses for you to check. An investigation can end with "No conclusion", and it says so.
  • It never turns an unmeasured value into zero.
  • When it cannot see, it tells you, instead of reporting healthy.
  • It records and proves. It does not block agent actions.
  • It gives no confidence scores.

INTEGRATIONS

What shoofi connects to

MODULE

AI agents

What shoofi is built for, on the same platform that watches the machines your agents run on.

Every agent run recorded with its tools, cost and outcome, plus an inventory of agents nobody has registered yet.

Included in Pro and Enterprise. An option on Starter.

Included in every plan

Linux hosts
CPU, memory, disk and network for every server or virtual machine.
Logs
Collected, searchable and tied to the host that produced them.
Traces
OpenTelemetry traces from your applications.
Alerts
Telegram, Slack and email.

Plugins

Starter includes 2 of your choice. Pro includes all of them.

Docker
CPU, memory, network and disk per container.
Kubernetes
Cluster, nodes and workload health.
Uptime & TLS
Checks from outside that your endpoints answer over HTTP and HTTPS, and alerts before a certificate expires.
PostgreSQL
Connections, queries, replication and storage.
MySQL
Connections, queries, replication and storage.
OpenSearch
Cluster health, indexing and search latency.

More coming soon.

Running something else? Tell us.

Off means off.

Every AI feature is switched on or off per workspace. When it is off, its routes do not exist for you, so there is no path to the model. And an investigation reads only your own workspace’s data.

PRICING

Three ways to run it

Prices in US dollars (USD), excluding VAT. The billing currency is agreed in the contract.

Starter
50 USD/month

Infrastructure monitoring

  • 5 supervised nodes
  • 10 GB of logs a month
  • 2 plugins of your choice
Pro
170 USD/month

Infrastructure and AI agents

  • 20 supervised nodes
  • 50 GB of logs a month
  • AI agent observability, 10 agents
Enterprise
On request

Deployed in your own infrastructure, including with no internet access

  • Limits per contract
  • Evidence chain verifiable offline
  • Investigations with your own LLM

See full pricing

Questions we get

What do I install?
An open-source collector on your side. We generate its config with your token. AI agents send their traces over OTLP/HTTP.
Which agent frameworks work today?
Proven end to end: LangGraph instrumented with Traceloop, over OTLP/HTTP. We validate other frameworks with each design partner.
How do I get access?
Accounts are created with an invite code. Book a call and we set you up.

Book a call

We onboard each design partner ourselves, on a call.