OpenTeams
Menu

Presented by OpenTeams

Own Your Intelligence

A Field Guide to the Distributed AI Economy

How AI capability becomes owned infrastructure — and what becomes tradeable when it does.

Read the white paper(PDF, opens in a new tab)

01 — The missing returns

Everyone is buying intelligence. Few are keeping it.

56%

of CEOs report zero financial impact from AI

Source

PwC, 29th Global CEO Survey, January 2026 (n=4,454 CEOs, 95 countries). Registry entry · as of 2026-07.

12%

of CEOs achieve both cost and revenue benefits from AI

Source

PwC, 29th Global CEO Survey, January 2026 (n=4,454 CEOs, 95 countries). Registry entry · as of 2026-07.

Corporate AI investment reached $581.7 billion in 2025, up 130% in a year, by Stanford HAI's count. Yet in PwC's survey of 4,454 chief executives, most report no financial impact from it at all.

The field guide's reading of that gap: it is not only a model-quality problem. It is anownership, context, integration, and accountability problem — an architecture problem. Better models alone will not close it.

Evidence current as of July 2026 — sources are listed with every figure.Registry

02 — Rented intelligence

The same spend. Two different balance sheets.

Consuming AI through systems you cannot inspect, reproduce, audit, or own has a compounding cost: every prompt that re-explains your organization teaches a system somebody else keeps.

Fig. 01 — Rented versus owned intelligenceAnimatedInteractive· 4 steps

The same organizational context either evaporates as an expense inside a vendor's system, or accretes as a versioned asset the organization keeps.

Rented

  • prompt
  • policy
  • style guide
  • process doc
  • correction

context re-explained, retained on the vendor’s side of the wall —expensed

Owned

  • prompt
  • policy
  • style guide
  • process doc
  • correction
company-contextv1 v2 v3

context captured, versioned, governed —capitalized

The organization is a tenant, and tenants do not get the keys to the building's mechanical room.
— the field guide, on vendor-hosted AI (§1)

03 — Context is capital

Your organization's context is an asset. Today it evaporates.

The rules, terminology, style, norms, and process knowledge that turn generic capability into your work are scattered across people, prompts, wikis, policies, and chat history — re-explained thousands of times a day, retained nowhere. The guide's aphorism:"Every prompt that re-explains the organization is capital spent as an expense."

Fig. 02 — Scattered context becomes a FrameAnimatedInteractive· 4 steps

Organizational knowledge scattered across wikis, prompts, policies and chat becomes a single scoped, versioned, governed Frame.

  • wiki page
  • prompt
  • style guide
  • Slack thread
  • policy PDF
  • onboarding doc
  • process checklist
  • brand rules
  • past correction

company-context.frame

sources:
9
owner:
knowledge-ops
scope:
organization
version:
v1.0.0

governed

Captured instead as a governed, versioned artifact, context stops being overhead and starts compounding. That artifact has a name — a Frame — and it is the first piece of the architecture.

04 — The architecture

From generic capability to accountable work

Five ideas, each building on the last. No product names yet — just the shape of the machine.

1 Frames carry context

A Frame is organizational context as a versioned artifact: scoped, inheritable, composable, shareable. A company Frame cascades into department and project Frames; a regulatory Frame composes in from outside. Context flows where it should — with an audit trail.

Fig. 03 — Frame inheritance and composition
A cascade of Frames inherited from company to department to team, joined by externally installed Frames for one working session Company Frame"who we are, how we speak, what we never do"Sales Department Framemethodology, stages, definitionsEnterprise Accounts Framesegment playbook, pricing rulesAn individual’s working session+Partner Frameinstalled from an external partner+Regulatory Frameinstalled from an industry consortium

A cascade of Frames. A Company Frame — who we are, how we speak, what we never do — is inherited by department Frames such as Engineering and Marketing, which are inherited in turn by team and project Frames. Separately installed external Frames, such as a regulatory Frame, compose alongside the internal cascade for a given piece of work. Context flows down the organizational structure with an auditable inheritance chain.

2 Cogs perform bounded work

A Cog is an AI worker you can hold to account: a model, oriented by Frames, with declared tools and permissions. Not a bare model — an assembly. Ask "what did this Cog do, under which Frames, with what outcome?" and the question has an answer.

Fig. 04 — Anatomy of a CogAnimatedInteractive· 6 steps

A Cog is an assembly of a model, Frame-founded context, declared tools, and governance parameters — the unit at which AI work becomes auditable.

A ring divided into four segments — model, context, tools, and governance — around a central core labelled doc-collector. Cogdoc-collector

cog: doc-collector v1.2.0

  1. modelpossibly specialized for the task
  2. contextfounded on one or more Frames
  3. tools & APIswhat it may call, and nothing else
  4. governancedata access, permitted actions, approvals

3 Ops package the job

Knowledge work comes in jobs, not tasks — "close the books," "run the quarterly compliance review." An Op packages a whole job: Cogs, workflow Frames, human checkpoints, integration logic, and a declared validation strategy, versioned and installable. "Notice what you cannot do with a job today: buy one."

4 Guards check. Gates decide. Tracks record.

A fluent output and a correct output look identical at a glance — and institutions that face auditors are not permitted to operate on glances. So verification is its own layer, and failure is part of the record.

Fig. 05 — Guards check, Gates decide, Tracks recordAnimatedInteractive· 6 steps

Guards verify the work, Gates decide what happens next, and the Track preserves the whole sequence — failures included.

draft findings
source-groundingpass
sensitive-datafailpass · rev 2
gaterevise proceed
approved output

Track — evidence, appended as it happens

  1. guard: source-grounding — pass
  2. guard: sensitive-data — FAIL
  3. gate: revise — returned to work
  4. guard: sensitive-data — pass (rev 2)
  5. gate: proceed

The failed check stays in the record. That is the point.

5 The Intelligence Hub keeps it yours

All of it runs inside an Intelligence Hub — a governed deployment within your own perimeter, where Frames live, Cogs run, Ops execute, and the evidence accumulates as Organizational Memory. Sovereignty here is realistic, not absolutist: control the model choice, the data paths, the policies, and the evidence — without manufacturing chips or training a frontier model.

Only now do the proper nouns matter: the Hub is assembled from the open-source commons —Nebari, the infrastructure stack, and Nebi, the packaging and environment layer. What exists versus what is proposed is trackedbelow, plainly.

Frames guide the work. Cogs perform the work. Ops orchestrate the work. Guards verify the work. Gates decide whether the work proceeds. Tracks make the work accountable.
— whitepaper, §5

05 — One job, end to end

Watch every abstraction become one job

The same example runs through this whole site: a quarterly compliance review, packaged as an Op. It is illustrative — no shipped product bears this name — but every term above has a seat in it.

Fig. 06 — Anatomy of an Op: quarterly-compliance-reviewAnimatedInteractive· 9 stepsillustrative example

An Op coordinates Cogs, Frames, Guards, Gates, and a human checkpoint into one accountable, installable job; a Track records everything that happened.

op: quarterly-compliance-reviewv2.3.1

company-policyregulatory-2026workflow Frames
doc-collector
policy-checker
exception-summarizer
compliance reviewer
permissionsrequired Framessource groundingsensitive dataschemaconsensusGuards check
  • proceed
  • request revision
  • require human review
  • escalate to expert
  • stop
delivered

supervising model: sequencing, retries, escalation · validation strategy: declared

Track — evidence, appended as it happens

  1. frames: company-policy, regulatory-2026
  2. cog: doc-collector — invoked
  3. cog: policy-checker — invoked
  4. cog: exception-summarizer — invoked
  5. guard: source-grounding — pass
  6. human: compliance reviewer — approved
  7. gate: proceed
  8. output: findings report — delivered

The guide's own self-test: cover the figure and explain why an Op written by a stranger can be safely installed in your organization. If your answer touches a manifest, a Frame, a Guard, and a Track — you have the architecture.Read it in full

06 — The whole stack

Drawn on one page

Fig. 07 — The whole stack on one page
The whole architecture drawn as one stack, with validation cutting across it Layer 3 — Economy§10–12The marketplace: Ops, Cogs, Frames and Guards published,discovered, installed across sovereign Hubs. Tracks stay home.Interface — The Desktop Application§9Where knowledge workers run Ops, converse with Cogs,and compose and share Frames.Layer 2 — Execution§3–5Ops — installable, supervised jobsCogs — AI workers oriented by FramesFrames — organizational context, versioned and inheritableLayer 1 — Infrastructure§7–8Organizational Memory — the context the Hub accumulatesIntelligence Hub — the governed deployment perimeterNebari — curated open-source stack · Nebi — packaging and installsCompute, open-weight models, the organization's dataValidation§6cross-cuttingGuards: checkGates: decideTracks: record

The architecture drawn as one stack. Layer 1, Infrastructure (Sections 7 and 8): the open-source commons assembled into an organization-controlled Intelligence Hub, with Organizational Memory accumulating under the organization’s governance. Layer 2, Execution (Sections 3 to 5): Frames carry context, Cogs perform bounded work, Ops orchestrate complete jobs. The Interface (Section 9): the application through which knowledge workers run Ops and converse with Cogs. Layer 3, Economy (Sections 10 to 12): artifacts exchanged across independent Hubs. Cross-cutting Validation (Section 6) spans every layer: Guards check, Gates decide, Tracks preserve evidence.

Infrastructure below, execution above it, the interface where people meet the system, an economy on top — and validation cutting across every layer.The guide keeps a legible copy

07 — The economy

Four kinds of artifact. One stays home.

Ops

trade like products — installable outcomes, priced like work

Cogs

trade like specialized labor — capability you can hire in

Frames

mostly circulate free — coordination goods that gain value by adoption

Guards

provide reusable trust — verification shared like test frameworks

Tracks

stay governed inside the organization. Evidence is not for sale.

Fig. 08 — Independent Hubs connect; Tracks stay homeAnimatedInteractive· 5 stepsillustrative exampleForecast or thesis

Independent, organization-owned Intelligence Hubs exchange Ops, Cogs, Frames and Guards without merging, while each Hub's evidence record stays local.

  • hub A · manufacturer

    • ops · cogs
    • frames · guards

    tracks: local

  • hub B · hospital network

    • ops · cogs
    • frames · guards

    tracks: local

  • hub C · consultancy

    • ops · cogs
    • frames · guards

    tracks: local

  • opfrom: hub-c · v1.1
  • cogfrom: hub-a · v2.0
  • framefrom: hub-b · v3.2
  • guardopen library · v0.9

what moves: ops, cogs, frames, guards · what stays: tracks

The most heavily exchanged artifact in this economy will probably never be sold.
— the field guide, on Frames (§10)

08 — Why open

The flywheel — with its brakes shown

Platforms tax exchange; protocols compound adoption. The bet is that open standards for Frames, Cogs, and Ops do for AI work what open packaging standards did for software. The guide shows the loop and the four ways it stops — because a flywheel is a forecast, not an achievement.

Fig. 09 — The network flywheel, with its brakes shown
A reinforcing loop in which more Hubs create a bigger publisher market, which creates more artifacts, which makes each Hub more valuable — and so on more Hubsbigger market for publishersmore artifactsmore value per Hub deploymentreinforcing loop

balancing forces, any of which can stall it

cold start
few Hubs, few publishers, little to install, weak case for Hubs
fragmentation
competing context and packaging standards split the network
quality collapse
a marketplace of mediocre Ops teaches buyers to stop looking
incumbent bundling
“good enough” closed features soak up the demand first

A reinforcing loop: more Hubs create a bigger market for publishers, which attracts more artifacts — Ops, Cogs, Frames, and Guards — which make Hubs more valuable, which attracts more Hubs. Four brakes can stop the flywheel: the cold-start problem (no artifacts, no buyers), fragmentation (competing incompatible standards), quality collapse (a marketplace flooded with bad artifacts), and incumbent bundling (platform vendors giving away adjacent capability). The flywheel is a forecast, not an achievement.

The guide states falsifiable indicators for whether this is working — its own, not the company's. Judge it by those

What exists, what is being built, and what remains a thesis

The guide keeps three tenses distinct, and so does this site: the infrastructure layer is real and inspectable now, the execution standards are hardening, and the economy is a forecast.

Status audited: 2026-07-25

Exists today

  • Nebari

    Exists today — open source, in transition to a modular stack

    Notes & source

    Open-source (BSD-3-Clause), latest release 2026.3.1 (July 2026). nebari.dev describes an ecosystem in transition: "Nebari classic" is sunsetting while "Nebari core" — the successor infrastructure layer — is labeled early access by the project itself. Say "production-proven" only of the classic platform; never call the successor GA.

    nebari.dev · 2026-07-25

  • Collab Desktop App

    Exists today — just released, free

    Notes & source

    openteams.com ships this as "Collab Desktop App — Just Released · Free": publish, discover, consume, and share agentic workflows and AI capabilities. The guide and whitepaper call this interface layer the "Desktop Application"; guide pages keep that canonical wording. No Web application is live as of the audit date.

    openteams.com · 2026-07-25

  • Intelligence Hub deployments

    Exists today — offered as an enterprise deployment

    Notes & source

    openteams.com offers the Intelligence Hub as its enterprise deployment tier. No deployment counts or customer numbers are published anywhere; this site must never state or imply one.

    openteams.com · 2026-07-25

  • Organizational Memory

    Exists today — a capability continuum, not a single product

    Notes & source

    Explicitly a continuum from versioned files to dedicated memory systems. "Many organizations will start here, and many will find this sufficient indefinitely." The safest component to describe publicly.

    whitepaper §3.4; field-guide §7 · 2026-07-25

In active development

  • Nebari core

    In active development — early access, per nebari.dev

    Notes & source

    The project's own label is "(early access)". Security, authentication, role-based access, and observability are design goals of the new core.

    nebari.dev · 2026-07-25

  • Nebi

    In active development — early release

    Notes & source

    nebi.nebari.dev says "early-release" (v0.13, Apache-2.0). Live scope today is environment management for teams (Pixi workspaces, sync/share, diff, OCI publishing). The whitepaper's larger role for Nebi — packaging and versioning Frames, Cogs, and Ops across Hubs — is roadmap, not current capability, and must be presented as such.

    nebi.nebari.dev · 2026-07-25

  • First Frames, Cogs, and Ops

    In active development — described by the paper as near-term deliverables

    Notes & source

    Every named artifact in the guide and on this site (including quarterly-compliance-review) is illustrative. No shipped, named Frame, Cog, or Op exists publicly as of the audit date.

    field-guide/02-field-guide.md FAQ 15 · 2026-07-25

Proposed standard

  • Frame protocol

    Proposed standard — publication is a near-term roadmap milestone

    Notes & source

    No published specification exists at any public URL as of the audit date. The guide itself says "once published, a near-term milestone on the paper's roadmap." The guide's own success indicator: a genuinely open specification with implementations not controlled by OpenTeams.

    field-guide/02-field-guide.md §3, §12; whitepaper §10 Phase 1 · 2026-07-25

Forecast or thesis

  • Public marketplace

    Forecast or thesis — a later roadmap phase

    Notes & source

    Not live anywhere. The guide places the public marketplace and vertical ecosystems in later phases; the nearest live analogue is nebari.dev's "Community hub (coming soon)".

    field-guide/02-field-guide.md FAQ 15; nebari.dev ("Community hub — coming soon") · 2026-07-25

  • The distributed AI economy

    Forecast or thesis — judged by the guide’s own falsifiable indicators

    Notes & source

    The guide's own framing keeps three tenses distinct: "the infrastructure layer is real and inspectable now, the execution standards are hardening, and the economy is a forecast." Section 12 lists falsifiable indicators — the guide's, not the company's.

    field-guide/02-field-guide.md §12, FAQ 15 · 2026-07-25

Go deeper

Why ownership mattersThe returns gap, and what renting intelligence actually costs.How the architecture worksFrames, Cogs, Ops — and the layer that checks the work.What economy it enablesFour kinds of artifact, exchanged across independent Hubs.
WhitepaperOpenTeams / Nebari PlatformThe Distributed AI EconomyIntelligence Hubs, Frames, Cogs, and OpsJuly 2026 · Revision 7

The Distributed AI Economy: Intelligence Hubs, Frames, Cogs, and Ops

OpenTeams / Nebari · July 2026 · Revision 7 · 35 pages

The primary source this guide interprets: the complete architecture — Frames, Cogs, Ops, the validation layer, the Intelligence Hub, and the economy thesis — in the company's own words.

Read the white paper(PDF, opens in a new tab)Download the white paper (PDF, 0.6 MB)

Sources & methodology ·Talk to OpenTeams ·Learn more about OpenTeams