Public beta now open

AI agents that can actually do the work.

AgenticSuite gives AI controlled access to your browser, computers, servers, and workflows, with ACT approvals for the actions that matter.

  1. Prompt
  2. Plan
  3. Approval
  4. Action
  5. Audit

Most AI agents stop at advice.AgenticSuite gives them controlled paths to act.

  • Your browser
  • Your computers
  • Your servers & VM hosts
  • Your workflows
  • Your agent runtimes

The suite

Three products. One control layer.

Let AI do the hard parts. Approve what matters.

The approval layer

ACT

Approve what matters.

ACT turns consequential agent actions into clear approval requests on your phone. Approve, modify, or deny before anything sensitive happens.

Signed, single-use approvals with a complete audit trail.

portal.example/renewal

Account verified

Form filled · 6 fields

Compared 3 plan options

Paused before checkout

The browser operator

BrowserBridge

Give AI your browser, not your trust.

Let AI browse, click, compare, fill, and complete web tasks in your real browser. It pauses before sensitive actions like submitting.

Runs in your own session. Panic control revokes in-flight authority.

$ ping pve01 · timeout

→ PiKVM console attached

observe: boot halted at initramfs

→ ACT approval · keys sent

recovered · summary written

The out-of-band operator

AgenticKVM

Operate beyond SSH.

Give agents a controlled path to operate computers, servers, and VM hosts when normal access is not enough. They use the same consoles an operator would.

Adapters for PiKVM, Redfish, IPMI, and BMC-style consoles.

Product demonstrations

Watch AgenticSuite work.

Three real-world workflows. One control layer.

Demo · BrowserBridge

Watch AI complete a browser task, safely.

“Find the earliest available appointment within 25 miles and fill the form, but stop before submitting.”

  1. Opens the browser and navigates to the site

  2. Searches and compares locations and times

  3. Fills the safe fields

  4. Pauses before submission

  5. You approve from your phoneACT approval

  6. BrowserBridge submits. Task done, audit recorded

Demo · AgenticKVM

Recover systems when normal access is gone.

“pve01 stopped responding after the kernel update. Diagnose from the console and recover safely. Ask before changing boot or power state.”

  1. Ping and SSH fail. Normal access is gone

  2. The agent opens a PiKVM/BMC console in observe mode

  3. It reads the boot failure and identifies the likely cause

  4. You approve the recovery action from your phoneACT approval

  5. Keys are sent, the machine boots, access is restored

  6. An incident summary is written to the audit trail

03:12:44Back online · summary written · every step recorded

Demo · ACT

Approve what matters from your phone.

Requests arrive as plain-English cards. You approve, modify, or deny. Every decision is recorded.

Requests route from every surface

  • Browser submission
  • KVM reboot
  • Hermes workflow
  • MCP tool action

Audit trail, written as it happens

  • 12:04:09Request received from BrowserBridge
  • 12:04:15Approved by you · single-use clearance issued
  • 12:04:17Clearance consumed · form submitted · task complete

How it works

How AgenticSuite turns prompts into controlled action.

Prompt. Plan. Approve. Act. Audit.

  1. Prompt

    Ask your agent to complete a task.

  2. Plan

    The agent decides what it needs to do across browsers, machines, or workflows.

  3. Approval

    ACT pauses consequential actions and asks for your approval.

  4. Action

    Approved actions run through BrowserBridge, AgenticKVM, Hermes, or MCP tools.

  5. Audit

    Every important request, decision, and action leaves a record.

For the technically curious
  • Scoped clearances: each approval covers one capability, not blanket access.
  • Signed approvals: decisions are cryptographically signed on your device.
  • One-time use: a clearance is consumed by the action it was issued for.
  • Provider-specific permissions: browser, console, and workflow surfaces each declare what they can do.
  • Fail-closed behavior: a missing or unverified clearance means no action.
  • Audit trail: requests, decisions, and actions are recorded for review.

Integrations

Built to connect with the tools agents already use.

AgenticSuite gives agent runtimes a controlled path into browsers, machines, and workflows.

  • HermesIntegration path for agent runtimes
  • Claude CodeDesigned for local agent runtimes
  • MCPFirst-class tool exposure where supported
  • BrowserBridgeChrome extension + local companion
  • PiKVMSupported console provider
  • RedfishAdapter for Redfish-compatible BMCs
  • IPMIAdapter for IPMI-compatible BMCs
  • BMC consolesWorks with compatible environments
  • iLOVia the Redfish/IPMI adapters
  • iDRACVia the Redfish/IPMI adapters
  • VM hostsConsole and power control where supported
  • GitHubSource, issues, and docs
  • TailscaleWorks with private tailnet-style networks
  • Local network toolsDesigned for local-first setups
  • Homelab infrastructureA natural home for the suite

Compatibility varies by environment and configuration. Product names belong to their owners; no partnership implied.

Security

Control is built in.

AgenticSuite is designed around human approval, scoped permissions, and visible records. Control is not an afterthought. It is the product.

  • Human approval

    Consequential actions pause for review.

  • Scoped access

    Agents get only the permission needed for the task.

  • Signed decisions

    Approvals are explicit and traceable.

  • One-time consumption

    Clearances are used for the action they were issued for.

  • Fail-closed defaults

    When control is uncertain, the system stops.

  • Observe before control

    Especially for KVM and infrastructure work, observation and action are clearly separated.

  • Audit trail

    Requests, decisions, and actions are recorded.

  • Panic and revoke

    Stop risky browser or system activity where supported.

Developers

Built for builders.

AgenticSuite exposes controlled action surfaces for agent runtimes, automations, and local-first infrastructure. Route browser actions, KVM operations, and workflow steps through ACT approvals and auditable execution paths.

  • Public beta
  • Self-hosted & local-first
  • MCP where supported
  • Open protocols on GitHub

# connect a runtime to controlled surfaces

agent → act(approval) # direct, or via hermes

  ├─ browserbridge · web tasks

  ├─ agentickvm · consoles & power

  └─ mcp tools · where supported

# every action → audit trail