Ask OLLIE. An AI worker that puts the right Skills to work.

Ask OLLIE doesn’t rely on one giant prompt. It combines reusable Global Skills with the live context and permissions of each client Workspace.

Give it the outcome. OLLIE selects the capabilities it needs, uses authorized tools, verifies the result and records what happened.

The Outloop assistant working at a desk

Ollie Runtime Pack 1.6.0-rc.4

Production candidate: package validated with 203 tests and S21 disagreements: 0. Desktop GUI installation and first-client live activation still need the documented smoke test.

Download the Ollie Runtime Pack ZIP · 320 KB
Installation guide Release notes Host compatibility First-client smoke test
Archive SHA-256 — verify before installing
0f35c2716a31de1b338638fb09450252a7767220d05dd270e2d9a4df7ef21c7f

Last updated:

One execution layer. Reusable capabilities. Each client’s own truth.

Ask OLLIE does the work
Runtime resolves authority and the Workspace. The Dispatcher selects eligible work and prevents duplicates; the Operator works toward the requested outcome.
Global Skills supply expertise
They are reusable capabilities, not rigid workflows or eight separate autonomous agents. OLLIE may use one, combine several, use relevant parts, or complete missing steps with other authorized methods. No perfect Skill is required to keep working.
Workspace Context supplies client facts
Identity, recipients, permissions, accounts, resources, business formulas, commission percentages and communication rules stay in the active Workspace and its tenant bindings.

“Analyze Meta performance and prepare a profitability report” can combine marketing analysis, Meta platform knowledge and profitability expertise. Runtime handles ownership and any authorized report or email delivery.

Browse the audited Skills catalog

From request to verified result

Architecture illustration — this explains the sequence; it is not a screenshot or live-run proof.

  1. 1.

    Request or scheduled intake

    Work arrives with a clear objective.

  2. 2.

    Resolve Workspace + authority

    Load live Context, permissions and resource bindings.

  3. 3.

    Dispatcher

    Identify eligible work and prevent duplicate execution.

  4. 4.

    Operator understands the task

    Choose Global Skills, combine them or use only the relevant parts.

  5. 5.

    Execute through authorized tools

    API / Outloop, permitted MCP, Browser / Computer Use or another authorized technique.

  6. 6.

    Verify

    Read back actual output and confirm the requested outcome.

  7. 7.

    Complete + report

    Deliver results, record evidence and name any remaining dependency.

  8. 8.

    Learn

    Reflective proposes reusable learning; approval updates one canonical Skill.

How Ollie works inside a client workspace

How work flows through Ollie and the Outloop workspace Work arrives from email, project management, a team request or a meeting follow-up. The Outloop workspace supplies identity, context, permissions, work state and audit. The Operator selects and combines Global Skills. Authorized tools include the Outloop API, permitted MCP and Browser / Computer Use or another technique. Results are verified by live read-back, and the team approves anything client-facing. 01 Inputs Email Gmail · Outlook Project management the intake queue Team request from a colleague Meeting follow-up an action that landed 02 Outloop Workspace Identity Live Context Goals & KPIs Rules Permissions Work state Approval policy Isolation Audit 03 Ollie Worker Understands the requested outcome Selects / combines Global Skills 04 Governed by the Outloop Workspace Outloop API preferred where the action exists Permitted MCP when authorized and appropriate Browser · other methods authorized recovery and UI work Client systems Files & media The mailbox 05 Verification Live read-back · evidence · duplicate prevention secret_exposed: false 06 Team control Task updated · internal draft prepared · final state recorded draft_for_internal_approval Approval before anything client-facing. How work flows through Ollie and the Outloop workspace 01 Inputs Email Project management + team request · meeting follow-up 02 Outloop Workspace Identity · Context · Permissions Work state · Approval policy · Audit Isolation · Goals & KPIs 03 Ollie Worker Selects Global Skills · combines relevant expertise 04 Governed by the Workspace Outloop API preferred where the action exists Permitted MCP when authorized and appropriate Browser · other methods authorized recovery and UI work Client systems · files · the mailbox 05 Verification live read-back · secret_exposed: false 06 Team control draft_for_internal_approval
  1. Inputs: email (Gmail or Outlook), project management, a team request, or a meeting follow-up.
  2. The Outloop Workspace supplies identity, live context, goals and KPIs, rules, permissions, work state, approval policy, tenant isolation and audit.
  3. The Ollie Worker understands the requested outcome, reuses complete or partial Skill logic, and plans the safest permitted path.
  4. The Operator selects Installed Skills as reusable expertise. Execution uses authorized tools: Outloop API, permitted MCP, Browser / Computer Use or another technique. Skills are not a transport.
  5. Verification: live read-back, evidence, duplicate prevention. secret_exposed: false.
  6. Team control: the task is updated, an internal draft is prepared, and approval is required before anything client-facing. The default policy is draft_for_internal_approval.

One Ollie operating system. Outloop gives every client workspace separate context, access, rules and work state.

The Operator selects Global Skills as expertise. Tools execute under Workspace authority. This is an architecture illustration, not evidence of a live run.

If one method fails, keep solving the task.

OLLIE is designed to complete the authorized task. A failed method does not by itself mean the task is blocked: METHOD_FAILED != TASK_BLOCKED.

Normal recovery follows API / Outloop → permitted MCP → appropriate Browser / Computer Use → another authorized technique → human escalation when genuinely required. Available routes depend on the live Workspace policy; an explicit denial is never bypassed.

These rules belong to Runtime’s canonical recovery owner. Business Skills consume them.

Email requests stay connected to the work.

One verified email thread stays linked to one PM task. OLLIE preserves the original Subject as the task title as closely as the PM system allows; later messages update that task.

The occurrence owner closes the loop with an internal teammate in the same thread. If work is incomplete, the reply explains what is blocked, exactly what the teammate needs to provide, and what OLLIE can still complete now. An authorized internal status reply does not require client-send approval.

If a message or file includes several clients, OLLIE uses only the active client’s confidently separated portion. If it cannot separate the information safely, it asks the internal sender for that client’s information separately, without copying other clients’ data.

An inconvenient Excel attachment is a reason to try practical authorized file, spreadsheet, conversion or browser methods before requesting another format. Readiness applies to each action: a pending client send does not cancel independent safe work or permitted internal communication.

Read the Email Lifecycle rules

Who decides what the client hears

Ollie does the work. The team keeps the client-facing boundary. These are two different permissions, and only the first one comes with the request.

The client-communication approval flow When work completes, Ollie asks whether client communication is required. If not, the internal task is updated and closed. If it is, Ollie prepares an internal draft which the team reviews. Changes revise the same draft. Only an approval opens the gate; the message is then sent in the original email thread, provider evidence is verified, and the task either waits for the client or closes. Work completed Client communication required? NO YES Update the internal task Nothing leaves the workspace CLOSED Internal draft prepared, not sent Team review Changes revise the same draft APPROVAL Send in the original thread Verify provider evidence WAITING_CLIENT CLOSED The client-communication approval flow Work completed Is client communication required? NO YES Update the internal task Internal draft prepared, not sent Team review APPROVAL Send in the original thread Verify provider evidence WAITING_CLIENT | CLOSED CLOSED
  1. Work completed.
  2. Is client communication required?
  3. If no: update the internal task; nothing leaves the workspace; the task closes.
  4. If yes: prepare an internal draft — prepared, not sent.
  5. The team reviews it. Requested changes revise the same draft, with no new thread and no new task.
  6. Only an approval opens the gate. Until then the draft cannot be sent.
  7. Send in the original email thread.
  8. Verify provider evidence.
  9. The task then waits for the client, or closes.

The draft reaches the gate and stops. Nothing crosses it without a named person.

Default policy

draft_for_internal_approval

Nothing seeds a standing approval. Approval is for the client-facing and high-risk boundary — not for every normal action Ollie is already permitted to take.

An authentic request authorizes Ollie to perform permitted work. It does not automatically authorize:

  • A client-facing send
  • Increased spend
  • An irreversible action
  • A legal or compliance exception
  • A policy bypass
  • Cross-workspace access

A successful API response is not proof a message was sent. A verified client send requires real provider evidence — a sent message id or its equivalent.

How the message itself is written

Whether a message may be sent and how it is written are two separate controls. Context decides the voice; the runtime decides the markup.

  • Voice comes from Context

    Outloop Context defines language, tone, signature, greeting and the client-specific communication rules.

  • Email-safe markup

    The runtime enforces structured, email-safe HTML rather than whatever a model happened to emit.

  • Hebrew

    Rendered right-to-left with lang="he".

  • English

    Rendered left-to-right with lang="en".

  • Mixed-language content

    Stays readable — a Latin product name inside a Hebrew sentence does not break the direction of either.

  • Proof of send

    Gmail and Outlook sends both require provider read-back evidence.

draft_for_internal_approval remains the default throughout. None of this rendering behaviour changes who approves a client-facing message.

Install once. Verify in the next client’s Workspace.

The ZIP contains eight complete Skills, installation and compatibility guides, tests, a manifest and checksums. It does not install Outloop, grant access or arm a schedule on import.

Runtime prerequisites — dependencies first

  1. 1. ask-ollie-authority-core
  2. 2. outloop-access-fallback
  3. 3. email-lifecycle-core
  4. 4. ask-ollie-workspace-bootstrapper

ask-ollie-authority-core

public-2.5-rc.4 · 2 files

Authority and permission boundaries.

Runtime Core · Candidate · local validation

Read ask-ollie-authority-core documentation
Component integrity

382cccd126f4be5513d1911e70ada5182583b5214708ec91206159fd905582f1

outloop-access-fallback

public-2.5-rc.3 · 2 files

Authorized recovery, route discovery and resource verification.

Runtime Core · Candidate · local validation

Read outloop-access-fallback documentation
Component integrity

50d32e5cd70b19d744d842430b768c50b2c9a8f766047a6f8fe990160af63e24

email-lifecycle-core

public-2.3-rc.4 · 5 files

Recipient identity, communication policy and stored-copy delivery evidence.

Runtime Core · Candidate · local validation

Read email-lifecycle-core documentation
Component integrity

5a2609f24e1d5d10a72bee9d8527e0f80bbcca124182d628b6652666205fb5ec

ask-ollie-workspace-bootstrapper

3.8.1-rc.4 · 24 files

Workspace bootstrap and generated Dispatcher / Operator.

Runtime Core · Candidate · local validation

Read ask-ollie-workspace-bootstrapper documentation
Component integrity

c3b7ae197c30a040cda5de347645927236ed1277e844bc934d024571879c3be3

reflective-skill-updater

2.2.0-rc.1 · 2 files

Propose approved, tenant-neutral learning to one canonical owner.

Runtime learning capability · Candidate · local validation

Read reflective-skill-updater documentation
Component integrity

c9a0337b51c6a84450b7e210e733fe8888daebc3e73453d267fb9638595e2f31

monthly-media-plan-status-report

1.1.0-rc.1 · 4 files

Compare planned media with measured delivery.

Optional Global Business Skill · Candidate · local validation

Read monthly-media-plan-status-report documentation
Component integrity

de6e192c3b3705864cb84f084b611a4d33810bb74881c534b91c85ddd4f63d7f

monthly-agency-profitability-report

1.1.0-rc.1 · 4 files

Calculate agency profitability from approved formulas and client inputs.

Optional Global Business Skill · Candidate · local validation

Read monthly-agency-profitability-report documentation
Component integrity

9b0fcc59b0c7652c2e6c76c496fd426e004258ffd698d2d8f66ded4dc6fe9901

crm-bmby-browser-reader

1.1.0-rc.3 · 4 files

Read Bmby with the approved Source Label binding.

Optional Global Platform Skill · Candidate · local validation

Read crm-bmby-browser-reader documentation
Component integrity

3ea97faa978b8dcecc3753b9a4bd88db292a2249029faea9eb785a0eb9580fb2

Claude Desktop / Cowork

Use the supported custom-plugin upload route for the whole ZIP. Local manifest validation passed for rc.4; CLI discovery of all eight Skills was recorded for rc.3 only. GUI import, enabled installation and the new client’s host execution remain unproven.

OpenAI / ChatGPT

Supported Work/Codex surfaces use extracted local or repository marketplace installation. Historical rc.3 CLI discovery returned an empty inventory; GUI import remains unproven. A normal cloud chat cannot reach the Local Bridge just because a Mac path is named.

Default schedule and approved deviations

Requested default: every two hours, 08:00–20:00, Sunday–Friday in the verified Workspace timezone. Cron shape: <mm> 8,10,12,14,16,18,20 * * 0-5. Resolve the minute from approved staggering and preserve explicitly authorized alternatives; verify saved timing and actual next runs.

A schedule may be armed only after the Workspace and host folder resolve, host access exists, intended schedule and scope are known, scheduling permission and capability are available, and duplicate-dispatcher checks pass. Preserve only explicitly authorized, runtime-verified deviations. Otherwise record the correct pending state and exact owner action.

An accepted Bridge request that is still running is pending work, not a missing connection. Runtime uses its supported claim / ownership / cycle / execution identity model. Client activation requires actual installed-byte read-back and the controlled and autonomous tests in SMOKE_TEST.md.

Inside Outloop

Real product screenshots from the existing documentation, cropped and redacted for public use. These show the Context, learning review and runtime health interfaces; they are not proof of an rc.4 installation or client run.

Outloop Memory and Context interface with organization sections collapsed and no contact details displayed
Context organizes the approved knowledge the agent uses. No client configuration is shown.
Outloop Agent Learning review header: approve, edit or reject before learning becomes memory; all workspace entries excluded
Learning is reviewed before adoption. The crop excludes all workspace entries and their contents.
Existing Outloop service health checks for broker, policy loading and audit writing, with the process identifier blacked out
Runtime health in an earlier product capture. This is status UI, not a new Local Bridge request receipt.

What has been proven

The 1.6.0-rc.4 package passed 203 tests and the S18–S22 consistency checks, with S21 disagreements: 0. The website serves the release owner’s unchanged ZIP. This proves package-level readiness; it does not prove GUI installation, a live provider send or first-client scheduled execution.

Read the proof and its limitsPrevious release — rollback only

Rollback keeps the prior release available at its original URL. Follow ROLLBACK.md; do not activate two dispatchers for the same work.

Ask OLLIE

Before you install

Ready to get out of the API loop?

Serve more clients with AI workers — and reuse what works.

Put AI workers to work across real client operations — and turn approved learning into reusable agency know-how.

Start 14-day trial Request a Workflow Review

For agencies and operators managing 5 to 100 client workspaces.