The desktop and application layer

Invocations and governance for humans, automations, and AI.

Work happens in applications: the CRM, the core system, the spreadsheet, the decades-old tool nobody wants to touch. InvokeFlow lets humans, automations, and AI agents operate that software directly, through each application’s own commands. No rewrites, no plugins, no screen-scraping.

Invocations, so they can do the work. Governance, so you can let them.

Get started See how it works
WindowsmacOSLinuxAny agent, any model
Who does the work
Humansat the deskAutomationsno model callAI agentsany model
Every command goes through
InvokeFlow
InvocationsEach application’s own commands, discovered from the running app
GovernanceEvery command checked where it lands, and kept on the record
Native commands, no plugins
Your applications, unmodified
CRMERPEmailSpreadsheetsCore systemsDecades-old apps
Who it’s for

Your software has three kinds of users now.

Humans at the desk. Automations running the routine. AI agents handling what needs reasoning. InvokeFlow puts all three on the same governed surface, and every action leaves a tamper-evident record.

Invocations · what they can do
Governance · what they may do

Humans

at the desk

Invocations

Open the job, not twelve windows: the right applications, arranged the way your best performer works, ready on command. Do the work once, and InvokeFlow can run it from then on.

Governance

The human stays in the seat and can take the controls at any moment. Destructive steps run only when a human names them, every run. What humans do is kept in the same record as everything else.

Automations

the routine, no model call

Invocations

Routine steps run deterministically, with no model call: faster, repeatable and free of token cost. Each step finds the exact live window again before it acts.

Governance

The same per-command checks, on every run, with each step kept on the record. There is no “approve all”: a destructive step still needs a human to name it.

AI agents

any agent, any model

Invocations

Any MCP-capable agent, on the model you chose, can call each application’s own commands, discovered from the running app. That includes software that never shipped an API.

Governance

Each command is checked against the exact application, process and window it is about to touch, at the moment it acts. Results go only where the grant allows. Overreach is refused, and the refusal is recorded.

How they work together

Each does the part it’s best at.

Keep humans in the loop where it matters. Automate what doesn’t need thinking. Use AI for what does. Most workflows need all three, so InvokeFlow runs them on one desktop, through one set of controls, with one record.

Example · a new lead lands

01 · AutomationThe routine, done

The CRM record opens, the routine fields fill and the dialer has the number. No model call.

02 · AI agentThe reading, done

An agent reads the account history and the call notes, and drafts the follow-up in the email client.

03 · HumanThe call, made

The rep reviews the draft, changes a line and sends it. The agent couldn’t have: sending is a destructive step, and no one named it for this run.

04 · The recordAll of it, kept

Every step, by all three, in one tamper-evident chain: which application, which command, under what authority, and how it ended.

See the record
Invocations

Further than the API goes.

An API is what a vendor chose to expose. The application can do more. Composer, InvokeFlow’s runtime for real desktop applications, calls what the software can already do, on Windows, macOS and Linux.

The whole callable surface

Every command the software already knows, discovered from the running application. No plugin, no vendor cooperation, nothing to rebuild when it updates.

An MCP face for every app

Discovered commands become tools any MCP-capable agent can call, giving software that never shipped an interface a working one.

Many apps in concert

Pull from one application, act in another, keep them in sync. Every command finds the exact live window again before it acts.

Teach it once

Do the work once and it can run it, in the same runtime, and prove it ran. It never records your screen to do it.

44% → 74% task success for the same AI model on the same desktop tasks, once it could invoke commands instead of finding and clicking. The model didn’t get smarter. It got a better way to reach the software. Wang, Li & Chen, EuroSys ’26. Why now

See the platform
Governance

Invocations without governance are a risk. Governance without invocations is a policy document.

InvokeFlow gives you both. Every command, from a human, an automation, or an AI agent, is checked against the window it is about to touch and kept on one tamper-evident record. You decide how far each one may go.

How governance works
Use cases

Start with the team that has the most windows open.

Sales & RevOps

Leads worked the moment they land, and every follow-up on the record.

Finance close

Balances pulled and matched across the ledger, the bank portal and the sheet.

Claims & underwriting

Every document for the case read and summarized, with the decision left to the adjuster.

Service desk

Tickets triaged and the fix drafted before the analyst opens them.

See all use cases
Start here

Start with one workflow.

Pick one that runs through the applications you already have. Humans, automations, and AI agents work it governed, and on the record, from the first command.

Get started How governance works