Sam
Get started
SAM FROM OPS / IN ACTION

What Sam shows you.
From the first call.

Connect a feature and Sam starts watching. Everything below is in the Sam dashboard today, for every team you run.

See the dashboard
WATCH IT RUN

Every call.
Every run, step by step.

Sam receives two small reports around each AI call: when it started, how it ended, the model or tool, tokens and timing. That’s enough to show you what your agents are doing.

Connect in minutes.

Create a key for a feature, copy a filled-in snippet in curl, Node or Python, and send a test event. The page shows it arrive, then sends a test loop so you see a finding before you write any code.

Live.

Calls stream in as they happen, model calls and tool calls side by side, with calls that started and never finished marked as they go stale.

Timeline.

Every step of every run on one chart, by when it started. Zoom from an hour to a week, filter by feature, and open any run to see its steps in order.

Activity.

Each finding with its evidence, runs that stalled, and the latest calls, filtered by feature and signal.

UNDERSTAND IT OVER TIME

What changed.
And where the work goes.

The same reports, added up by day, by feature, by model and by tool.

Analytics.

Tokens, calls and runs over time; which models and features take the most; latency by model; error rates; when in the week your agents are busiest; and a short list of what stands out.

Insights.

Every model each feature calls, and how that’s changing. Models with an announced retirement date are flagged, with the features still calling them.

Tools.

The tools your agents call, how often, how long they take, how often they fail, and which step tends to follow which.

Usage.

Agent steps this month against your plan’s allowance, on Billing and Analytics. Going over never stops Sam watching.

WHEN SAM IS WRONG

Sam will get some wrong.
Here’s what you can do.

A teammate earns trust by what happens after a mistake. All of this works today.

Tell Sam it was wrong.

Every finding has a “This was wrong” button. It’s recorded with the finding, and we measure Sam’s precision from those marks, not from how many findings it raised.

Change the limit.

A research agent is meant to explore. Raise a check’s limit or switch it off for the features where it doesn’t fit, and keep it everywhere else.

Halting is your switch.

New teams start with Sam observing. Switch halting on and a finding returns a halt to your agent on its next call; switch it off and none can. Every change is recorded.

A halt is a request.

Your agent decides what to do with a halt. It’s a request your code must honour, not a stop Sam forces from outside.

FOR THE WHOLE TEAM

One organization.
Everyone who needs to see.

Invite people to your organization in Sam. On Custom, run several teams, each with its own keys, findings and settings.

People and roles.

An owner, admins and members. On Pro, give members access to just the teams they work on, to manage or only to view.

A record of every change.

Each team and the organization keep an audit log: keys created and revoked, halting switched, people invited, roles changed.

SMALL INTEGRATION. ONGOING ATTENTION.

Two calls around yours.
No prompts handed over.

Report the shape of each AI call without waiting on Sam. Keep your application’s content where it belongs.

See the integration