For multi-client agencies & GTM teams
More client work.
Same team.
Run AI workers across every client’s accounts, files and systems — without mixing clients or rebuilding setup every time.
Workspace infrastructure for real AI workers.
For your company-controlled Mac. Setup guide
Each client gets its own isolated Workspace.
Google Ads and Meta requests for the client’s own accounts run. Requests for another account are refused.
What this demo shows
- This Workspace’s accountRequest succeeded
- Wrong accountRequest refused
- Unapproved campaignRequest refused
Read the log details
- RESOURCE_ID_NOT_ALLOWED
- Google Ads or Meta account outside this Workspace’s approved resources.
- META_OBJECT_NOT_APPROVED
- Meta campaign, ad set or ad outside the approved account. The action was not sent to Meta.
- Allow · OK
- Approved requests returned results in the same Workspace.
Demo recreation, not a customer performance claim. Example tool logos belong to their owners.
See the client work behind the workflow.
A website rebuild. A missing conversion step. A team directing the work. Adam walks through one real client project with an AI worker and Outloop.
Watch the founder case study
Reported timings; no measured advertising uplift.
Reuse the method. Keep each client separate.
Reuse reviewed workflows. Configure each client’s Context and validate their accounts and resources separately.
Explore repeatable implementation →Concept illustration, not measured gains. Example tool logos belong to their owners.
-
More delivery capacity
-
Reusable methods
-
Less repeated setup
-
Separate client data
Choose a supported AI runtime
Models change. The Workspace stays.
Keep client Context, Skills, approved access and work history as you switch supported AI runtimes.
Runtime support and limits
In Stable 1.45.0, Routines keep instructions, schedule and history when switching between ready Claude Code and Codex runtimes. Claude Code, Codex and Hermes differ in setup and feature support. Full conversation, running-task and cross-machine migration are not promised. Private lessons do not automatically move between clients.
Open/local options depend on support. Outloop does not route models automatically or guarantee savings. Compare runtime economics →
Read the video transcript
Imagine giving your AI worker a real workplace. A place where it has the right files, instructions, business tools, and approved access to get work done. That's what Outloop is built for. Instead of starting from scratch with every task, your AI worker works inside a dedicated Workspace. It knows which client it's working for, which resources it can use, and where its work belongs. And with Ollie, you can ask for real work, not just suggestions. Analyze campaigns. Prepare reports. Work with client systems. Handle recurring tasks. Your team stays in control. The AI worker handles the execution. More client work. Same team. That's Outloop.
Download English captionsRecorded in Outloop 1.45.2. Campaign figures are synthetic demo data, not live advertising results.
Independent tools. Names and logos belong to their respective owners; Outloop is not affiliated with or endorsed by these projects.
The anatomy of a real AI worker
The model provides the intelligence. Ollie does the work. Outloop is the harness that connects everything — with each client’s knowledge, accounts and access kept separate.
Layer 1 of 5
The brain provides intelligence.
A model does the reasoning. A runtime — Claude Code, OpenAI Codex, or Hermes running a local model — turns that reasoning into tool use. Choose the runtime that fits the work; the client Workspace stays the same.
Layer 2 of 5
Meet Ollie. Your AI worker.
Ollie receives requests, works with the client’s context and approved capabilities, prepares results for review and executes authorized tasks.
Layer 3 of 5
Outloop is the harness.
The infrastructure that lets AI workers operate across real business systems with approved access and controlled execution: access and permissions, context and skills, browser and API routing, human approvals, continuity and audit.
Layer 4 of 5
Real access. Real execution.
Ollie works through the Managed Browser with Browser Login recovery, any REST API through the Universal API Bridge, the MCP servers you already use, your own servers over SSH, local files and folders, and business apps such as Google Ads, Meta, Google Drive, Gmail and Microsoft 365 mail — with credentials kept out of the AI’s view.
Layer 5 of 5
One infrastructure. Separate client Workspaces.
Scale client work while keeping each client’s knowledge, accounts, permissions and execution separate. A request for another client’s account is refused before anything is sent.
The brain
Model = reasoning · Runtime = where it works

OpenAI
Local model
Outloop · the harness
- Access & PermissionsEach client’s approved accounts only
- Context & SkillsClient knowledge and reviewed methods
- Browser & API RoutingBrowser, API, MCP and SSH paths
Ready for client work
OllieYour AI worker
- Human ApprovalsA person approves what matters
- Continuity & AuditRoutines, Learning and a record
Real execution
- Google Ads
- Meta
- Google Drive
- Gmail
- Google Sheets
- GA4
- Browser Login
- API Bridge
- MCP tools
- SSH servers
- Files & Mac
- Business apps
Isolated client Workspaces
-
Client AWorkspace
- Knowledge
- Skills
- Accounts
- Files
- Permissions
- History
Own Google Ads account
✓Weekly report returned
×Client B account refused · nothing sent
-
Client BWorkspace
- Knowledge
- Skills
- Accounts
- Files
- Permissions
- History
Own Meta ad account
✓Campaign draft returned
-
Client CWorkspace
- Knowledge
- Skills
- Accounts
- Files
- Permissions
- History
Own Google Drive
✓Document analysis returned
Diagram. Layer 1, the brain: a model, run in Claude Code, OpenAI Codex or Hermes with a local model. Layer 2: Ollie, the AI worker. Layer 3: Outloop, the harness around Ollie, with Access and Permissions, Context and Skills, Browser and API Routing, Human Approvals, and Continuity and Audit. Layer 4: real execution through Google Ads, Meta, Google Drive, Gmail, Google Sheets, GA4, Browser Login, the API Bridge, MCP tools, SSH servers, files and business apps. Layer 5: three separate client Workspaces. Client A gets a Google Ads weekly report, Client B a Meta campaign draft and Client C a Google Drive document analysis, each returned to its own Workspace. When Client A asks to use Client B's Meta account, the harness refuses before anything is sent.
Execution and browser details
- Local ChromeUses the Workspace’s configured profile.
- Managed BrowserShared profile; supported login recovery stays on the task page.
- Human handoffsPause for authentication or review; verify the result.
Local files, installed apps and your network or VPN are available where configured. Support depends on the service, runtime and configuration.
Explore execution and browser modes →Interactive illustration. Example tools; logos belong to their owners. Client names are illustrative. Independent tools. Names and logos belong to their respective owners; Outloop is not affiliated with or endorsed by these projects.
See a marketing workflow in action.
Watch a client brief become reviewed Google Ads and Meta campaign builds.
Demo inputs; real builds, PAUSED. Review emails are staged reconstructions. No ad delivery or performance claim.
67-second Ellātu preview · Google Ads + Meta · No form required
Read the video transcript
Watch an AI worker take one agency job from the client brief all the way into Google Ads and Meta Ads. It starts with a brief, in the right workspace, with the client's approved Context. The worker is Ollie, running inside an Outloop workspace with the client Context, approved access and approval rules it needs to do the job. Ollie researches the market, plans the campaign structure, and builds the creative from the client's real Drive assets. Then the Account Manager reviews the package, and changes two things. After approval, Ollie doesn't stop at a recommendation. It builds the campaign, pushes it into Google Ads and Meta Ads, keeps everything PAUSED, and reads the build back through the API. That was only part of the workflow.
Download English captionsYour first workspace, in three steps.
- Create a client workspace.
- Add Context and approved tools.
- Give your worker a task.
Next: Ask Ollie to do real work →
0:57 · Adam’s voice · English captions
Recorded in Outloop 1.45.2. “Create a Workspace” is an animated explanation; the creation flow is not shown.
Read the video transcript
Let's set up your first AI worker in Outloop. Everything starts with a Workspace. Think of it as your AI worker's permanent working environment. If you run an agency, you might create one Workspace for each client. Each Workspace has its own folder, context, skills, and approved access. First, create or connect the folder where the work belongs. Next, add the information your AI worker needs to understand the client and the job. Then, connect the business services it needs, and make sure the right access is assigned. Once your AI runtime is ready, you can start giving Ollie real tasks. The important thing is that your Workspace remains in place between tasks. You don't need to explain the same client setup every single time. Your AI worker has a place to work. And you're ready to get started.
Download English captions
Built by an agency operator
I'm Adam Argaman. I built Outloop for the client work my agency team does.
The right client Context. Approved tools. Human review.
Why we built Outloop →Start with one real client workflow.
Choose one recurring job to try. Follow the first-Workspace guide →
Pro starts at $99/month with 2 Workspaces included. Start your 14-day trial on Outloop Cloud and review the final price at checkout.
Need help? Request an AI Workflow Review.
Send us one real agent workflow. We'll show you where access will break.
A 15-minute AI Workflow Review of one workflow you actually run. We’ll review one real workflow from your agency, identify the tools, accounts, files and permissions it requires, and show you how it could become an AI worker workflow.
Ready to start right away? Start your 14-day trial → It begins on Outloop Cloud (account & license only); your API keys stay local on your Mac.
✓ We never ask you to paste secrets into this form.
In the review, we map:
Built for agent workspaces like:
Independent tools. Names and logos belong to their respective owners; no official integration is claimed.
Request your 15-minute AI Workflow Review
We review your workflow · We reply by email · No card needed to talk to us
Or start the 14-day trial nowSecrets stay local · We never ask you to paste secrets into this form.
Do not paste API keys, passwords, Bearer tokens, Authorization headers, .env contents, or other secrets into this form.
By submitting, you agree that we may contact you about your Outloop workflow review. Terms · Privacy · Refund Policy
Request received.
We will review your workflow and contact you at about your workflow review. Do not send API keys or secrets by email.