Every AI agent and every run, in one console.

ControlTower is the admin console for AI agent governance: administrators see every agent, its budget and its access, and a record of every run. Below, a single run, line by line.

Book a demo See an agent at work

AI agent governance starts before the first run.

Nothing to bolt on before the first run: a new agent already has a budget, an access list, your privacy rules, an approved‑model list and a record, and it can be paused. Each rule, except the pause, writes one line on the run's record, shown next.

  1. Budget Its sponsor's (the person or team that funds it), or the caller's. A run with no budget does not start. on the record COST
  2. Access Only the tools and data it was given. on the record ACCESS
  3. Guardrails Personal data and secrets masked by the rules you set. on the record PRIVACY
  4. Models Through SecondGate, the LLM gateway, from the models its team approved. on the record AI MODEL
  5. Records Every run on a record administrators can read, not change; conversations stay in your own database. on the record RECORD
  6. Pause Its owner or an admin can pause it, any time. in ControlTower

Follow one run, line by line.

An ordinary run: the named agent @finance-⁠analyst writing its Monday vendor report. Its record has eight lines. The two ControlTower owns, cost and record, open below; the other six open under Agents and Gateway.

COST · $0.31 → Finance team

Every run lands on a budget someone owns.

A named agent is charged to its sponsor, the person or team that funds it, or to whoever called it; your personal agent to you. Alerts go out when spend passes a threshold you set.

Budgets · Octoberthis month
Organization $3,912.40 / $12,000
Finance team $184.20 / $600
@finance-analyst sponsor: Finance $4.96 this month
RECORD · kept in Agent Activity

Trace any run to its sources and cost.

In Agent Activity, ControlTower's log of runs, administrators follow a run from any channel down to what it asked the AI model, what it read, where the result went and what it cost. Administrators can read the record but not edit it. It holds what ran; the chat itself stays in your own database.

ControlTower · Agent Activitythis week · 212 runs
any agent any channel last 7 days
AGENTWHERECOSTSTATUS
@finance-analyst Schedule $0.31 Done
Personal agent · m.kim Slack DM $0.08 Done
@legal-adviser Telegram $0.92 Done
Project agent · Q4 vendor review Web app $0.03 Done
@tech-writer Slack bot $0.14 Issue
@finance-analyst · Monday 07:00 run · started by its schedule
01Asked the AI model to plan the report$0.19
02Read the vendor sheet in Google Sheets · 2 values masked—
03Asked the AI model to write the summary$0.12
04Posted to #finance-⁠ops in Slack · message + file—

Every named agent on one list.

The Agents list in ControlTower shows every named agent: owner, budget for the month, last run and anything still missing from the set‑up. The pause button is here.

ControlTower · Agents3 named agents
AGENTOWNERTHIS MONTHLAST RUN
@finance-analyst Finance $4.96 of $50 Mon 07:00 · completed
@legal-adviser Legal $21.40 of $80 today 10:12 · completed
@tech-writer Engineering $0.00 of $40 Slack bot not connected
@finance-analystpublished · owner Finance · activePause

Questions administrators ask.

What does ControlTower record?
Agent Activity records each run: what started it, its model and tool calls, where the result went and what it cost, with a trace reference and credentials stripped out. Administrators can read it, not change it. Conversations stay in your own database, and a deleted conversation leaves no copy in the record. Guardrail findings are logged too; your administrators choose whether they keep no text, masked text or the original.
Who decides which models an agent may use?
Your administrators, in ControlTower. Which models a team, a person or an agent may use is a list they set, from the providers where you already have accounts, including Anthropic, OpenAI, Google and self‑hosted models, on your own keys. We do not host or resell models.
Agents Personal, project and named agents: how they start, what they may use, how you hire one. Gateway One door to every AI model, with the checks, cost control and masking inside it.

See one run, from request to record.

On the demo call we follow one run the way an administrator sees it in ControlTower: who owns the agent, which budget paid and what the record keeps.

See how SecondGate checks every AI call 
Book a demo