The product
Deploy AI workers into real company work.
More client work. Same team. Give each client a durable Workspace for Context, approved accounts and resources, reusable methods and real execution. Your company-controlled Mac is the execution environment; the Workspace is the operating home.
Last updated:
In short
Outloop is Workspace infrastructure for deploying and operating AI workers.
Outloop provides Workspace infrastructure for deploying and operating AI workers in separate client Workspaces, with client-specific Context, approved access, resources and reusable methods. It runs on your company-controlled Mac today. Your model, runtime and provider accounts remain separate choices.
Reuse approved API access across the runtimes your team already uses
One approved access layer for Cowork-style sandboxes, Claude Code, Codex, Hermes, and OpenClaw — without rebuilding setup for every platform. Add or change an approved runtime; context and approved access stay attached to the client workspace.
One credential. The right workspace. Any approved agent runtime. secret_exposed:false
Independent tools. Names and logos belong to their respective owners; Outloop is not affiliated with or endorsed by these projects.
One client Workspace. The right approved action.
Follow a supported request for the intended client. An out-of-scope request is refused; it is not permission to switch tools and bypass the boundary.
One request, from ask to audit
- Ollie starts the intended client task.
- Use the Workspace’s approved accounts and resources.
- Supported approved work returns a result.
- A wrong-client request is refused.
- Review the result and work record.
Ollie does the work. Outloop controls how the work happens.
Concept animation of client-bound work. Approved actions use the configured client resources; out-of-scope requests are refused. Check the recorded Google Ads and Meta evidence below for specific proven behavior.
Ask Ollie to do real work.
Start with one task in the right Workspace. Ollie works with its available information, files and approved tools, then returns work you can inspect and follow up on.
This recording follows a campaign-analysis request through to a result and a follow-up. The figures are synthetic; they demonstrate the workflow, not client performance.
A completion report is a starting point for review. “Reported complete” does not mean “verified by Outloop.” Check the output and its supporting evidence before relying on it.
Prepare your Workspace → Next: make repeat work a Routine →0:56 · Adam’s voice · English captions
Synthetic demo data. The roughly one-minute wait is shortened and labeled. “Human approval” is an animated explanation, not a recording of two people reviewing an action.
Read the video transcript
Now that your Workspace is ready, let's put Ollie to work. Imagine you're managing marketing for several clients. Instead of opening different dashboards, downloading reports, and analyzing everything manually, you can ask Ollie to do the work. For example: Review this week's campaign performance, compare the results against our targets, and tell me what needs attention. Ollie can work with the information, files, and approved tools available in that Workspace. It can calculate metrics like return on ad spend, identify performance changes, and prepare recommendations. You can inspect the result and ask follow- up questions. And when a workflow requires human approval, Outloop provides a way to review the proposed action. The goal is simple. You focus on the decisions. Your AI worker handles the repetitive execution.
Download English captionsDefine the work once. Run it again.
A Routine saves a task inside one Workspace. Describe the work, choose a ready AI engine, and select a manual run, a schedule or a supported event trigger.
Use the execution history to inspect what ran and what the worker reported. Review the returned work and any action that requires human approval.
The recording shows the Trigger menu. Its approval card explains the review step; it does not demonstrate a two-person approval flow.
Connect the tools your task needs → See the real scheduled runs →1:10 · Adam’s voice · English captions
Synthetic demo data. The Trigger menu is recorded product UI. “Human approval” is an animated explanation, not a recording of two people reviewing an action.
Read the video transcript
Some work needs to happen again and again. Daily reports. Weekly performance reviews. Competitive research. Regular client updates. You shouldn't have to give your AI worker the same instructions every time. That's why Outloop has Routines. A Routine is a saved task that runs inside a specific Workspace. You describe the work, choose the AI engine, and decide when it should run. You can start it manually, schedule it, or use a supported event trigger. For example, you could ask Ollie to prepare a marketing performance brief every morning. It can review the available campaign data, calculate return on ad spend, identify important changes, and prepare recommendations. Outloop records the execution history, so you can inspect what happened and what the worker reported. And when a workflow needs human approval, you can review the relevant action before it goes ahead. Define the workflow once. Let your AI worker handle the recurring work. And spend your time on the decisions that matter.
Download English captionsSee the real scheduled runs.
Inspect saved Routine settings and two timer-triggered executions. The footage shows scheduled times, run history and returned results. The marketing figures are synthetic demo data.
Watch the Routines pilot · 1:09
1:09 · Adam’s voice · English captions
Real scheduled executions with synthetic figures. “Reported complete” is not “verified by Outloop”; inspect the returned work before relying on it.
Read the video transcript
A Routine is work your AI worker does on a schedule, inside one Workspace. These are the saved settings of a Routine called Marketing Brief. The instructions describe the task, using synthetic figures. It belongs to this Workspace only. It runs with Claude Code. And it was set to run daily at five fifty- six PM At five fifty-six, the timer fired, and the run finished in thirty-one seconds. Outloop shows the result, and labels it honestly: reported complete, not verified by Outloop. The figures are synthetic demo data: Search and Social totals, a blended return on ad spend, and two recommendations. This is a different Routine, from earlier that day: a weekly marketing summary, scheduled for five forty. Its run took twenty seconds, with one attempt, and Outloop records when it was scheduled, when it started, and when it ended. Define the task, choose the engine and the schedule, then review what the scheduled run returned. That's a Routine in Outloop.
Download English captionsWatch the request flow run in the product.
The layer described above, recorded: a worker requests approved access inside its client workspace, the action executes, and requests outside the workspace are refused and logged — the credential stays in Outloop throughout.
- Workspace-scoped action requests
- Labeled dashboard views and recorded denials
- Redacted API evidence
Create your trial, then install on your company-controlled Mac. Set up your first workspace.
56 seconds · One request, the right client · No form required
The opening is an architecture illustration. Dashboard views and recorded denials are labeled in the video; API evidence is redacted. Shown campaign execution stays PAUSED.
See the actual client boundary
Outloop’s Live Activity from one demo workspace running Google Ads and Meta Ads work with approved client access. Approved requests execute; requests for an unapproved resource are refused.
Swipe sideways to read the log · open full size
Refused: these wrong-resource requests did not run
- Deny ·
RESOURCE_ID_NOT_ALLOWED(google_ads and meta_ads) — the request named an account outside this Workspace’s approved resources; the requested action was refused - Deny ·
META_OBJECT_NOT_APPROVED(meta_ads) — a campaign, ad set or ad that does not belong to the approved ad account; the requested action is not sent to Meta - Approved requests in the same workspace keep executing around the refusals — Allow · OK
Real execution in an Outloop demo workspace (workspace 018), published in the September 26, 2026 proof pack — not a customer case study, and not evidence of ad spend, delivery or performance. Enforcement covers requests routed through Outloop. See the runtime access walkthrough for the recorded flow.
The eight layers, and what each one is for.
Each layer carries its own evidence label. Where something is built but not yet broadly proven, it says so. Two things are deliberately not on this list — the model and the agent runtime. Both are yours to choose and to change. The eight layers are what stays.
- 01
A durable client Workspace
current supported scopeClient identity, Context, Skills, approved resources, Routines and work records stay organized around the work. A client, project or department can be the operating boundary.
Models change. The Workspace stays.
Explore this layer - 02
Approved Context and reviewed Learning
current supported scopeKeep each client’s brief, working rules and approved lessons with that Workspace. Knowledge does not grant additional access.
Review changes before they become working knowledge.
Explore this layer - 03
Client isolation
current supported scopeEach client has its own Context, approved accounts/resources, work history and private state.
The right answer must also reach the right client.
Explore this layer - 04
Reusable implementation
current supported scopeReuse reviewed client-neutral integration setup, workflow logic, Skills and Routine patterns. Configure and validate each new Workspace separately.
Qualitatively proven reuse; no universal setup-time benchmark.
Explore this layer - 05
Supported runtime choice
current supported scopeStable 1.45.0 Routines retain instructions, schedule and history between ready Claude Code and Codex runtimes. Other workflows have their own support limits.
Evaluate quality and cost without redefining the client. No automatic routing or universal session migration.
Explore this layer - 06
Real execution
current supported scopeSupported APIs, browser work, files and installed applications can participate in client workflows, with authentication and human review when needed.
Local Chrome uses the configured Workspace profile; Managed Browser uses a shared profile and supported same-task login recovery.
Explore this layer - 07
Approved accounts and resources
current supported scopeRecorded Google Ads and Meta examples show approved requests succeeding and out-of-scope requests refused through Outloop.
Controls cover configured supported actions, not unrelated tools.
Explore this layer - 08
Work records and review
current supported scopeKeep records and outputs with the client Workspace. Inspect the result and review consequential decisions before continuing.
Private client state is not a reusable template.
Explore this layer
What Outloop is not.
- The AI worker itself — Outloop supplies neither the model nor the agent runtime nor the computer. You choose those.
- A general agent framework or project-management system.
- A cloud agent host — the agent runs on your machine, not ours.
- An MCP marketplace or an MCP manager.
- A campaign automation or reporting tool — the agent does the work; Outloop controls the access.
- A replacement for 1Password, Infisical, Doppler or macOS Keychain.
What agents actually do with it: the AI agent capability catalog
· ready-to-download agent Skills.
More on the boundaries: Outloop is not a vault
· the Workspace explained
· the long-term thesis.
What’s included — and what you bring.
Outloop does not include a computer or an AI model. It is the Workspace infrastructure around the AI worker you run, on a Mac your company controls.
Included in your plan
- The Outloop Mac app: client workspaces, approved API access, the managed browser with Browser Login recovery, file access, Context and the redacted audit.
- Separate client Workspaces, approved resource configuration and work records.
- Setup guides for supported services and Custom API configuration, plus product support.
What you bring
- A company-controlled Mac to run it on. Outloop does not sell, host or provide the computer.
- Your AI model and agent runtime (for example Claude Code or Cowork) — Outloop supplies neither, and a cloud model may process task content under its own terms.
- The provider accounts and API access your clients approve, plus any provider costs.
- Custom implementation / FDE work, if you want it, is priced separately.
Plan prices: pricing · estimate implementation separately with the implementation cost calculator.
Choose the engine around the Workspace.
Models change. The Workspace stays. Supported tools such as Claude Code, Codex and Hermes can work around the same client identity, approved Context and access. Other supported environments have their own setup and capability limits.
Approved runtime your team uses
Working in: your approved runtime
- Client context
- Approved access
- Reviewed learning
- Client boundary
- APIs
- Browser logins
- Files
- Client systems
Let the economics change without redefining the client.
Use a lower-cost supported model or runtime when it meets the workload’s quality and tool requirements. Use a frontier model where its capabilities justify the cost. The Workspace keeps the client identity, Context, Skills, approved access and durable work organized around supported runtime choices.
That makes economic optionality more useful than portability alone: when a more economical supported option fits the job, the client’s operating setup remains the starting point. Current Routine continuity is specific: instructions, schedule and history remain when switching between ready Claude Code and Codex runtimes. Smaller, open and local models are a strategic opportunity as capability and integration support develop, not a promise of universal support or parity today.
Evaluate cost per completed workload, including review, retries and operations—not just model pricing.
Outloop does not choose or route models for you. A runtime change does not promise conversation transfer, resumption of an in-progress task, or complete migration to another machine or cloud. Check current capabilities.
Use the surface the work needs.
Client work can span approved APIs, persistent Chrome sessions, files and installed Mac applications. Company network or VPN access depends on the configured machine. A supported API may handle structured actions; a website may expose a task unavailable through that API; a local file may be the input or final deliverable.
The client and approved access remain the starting point. A denied action is not permission to bypass the boundary through another tool.
Browser identity depends on the configured mode.
Local Chrome: a Workspace uses its configured Chrome profile for supported browser work and Browser Login. Separate configured profiles provide separate browser environments.
Managed Browser: supported recovery can continue on the same task page using an approved authenticated session. This mode uses a shared profile, not a separate profile per Workspace. Approved Workspace accounts and resource boundaries still apply to supported actions through Outloop.
Human handoffs are part of the workflow.
Required review and authentication can pause work. Browser Login recovers supported, configured sign-ins. Installed QA records successful recovery and continuation in the same task page. Some challenges still need human attention. Inspect the actual completion state before retrying an interrupted action.
See documented website work · See a creative workflow · Compare API and browser execution.
Keep useful context. Review what changes.
Ready to start using your worker? Follow the Ollie guide: built-in Skills, first requests, email and approvals.
Approved Context gives the worker the client’s brief and working rules. Proposed lessons remain subject to human review. Adding knowledge does not expand access, and learning for one client is not automatically applied to another.
Use Context and reviewed learning to improve the client’s working setup. Reuse only reviewed, client-neutral methods across the team.
Keep your vault. Control runtime access.
Today Outloop keeps approved credentials in macOS Keychain on your company-controlled Mac. Direct integrations with external vaults such as 1Password, Infisical and Doppler are not supported yet. Outloop never replaces your vault — it controls which workspace and runtime can use approved access.
- No API keys uploaded to cloud.
- No raw key returned to the agent.
- No .env files required.
- Wrong-client access is blocked before credential use.
See it on one of your own client workflows.
Start the 14-day trial and run your first approved access proof locally — or send us one workflow and we will show you where the access breaks.