Additional

CoWork OS Positioning and Go-to-Market Strategy

Synced from github.com/CoWork-OS/CoWork-OS/docs

Last reviewed: 2026-08-27

Canonical Positioning

CoWork OS is the free, open-source AI super app for real work. It brings coding, research, browser work, inbox, documents, spreadsheets, presentations, channels, automations, and agents into one GUI-first, CLI-capable workspace.

Underneath that super app is an open, multi-provider agent harness. Users can connect supported provider accounts, API keys, compatible gateways, cloud credentials, or local models and keep the same CoWork tools, memory, agents, approvals, artifacts, and workflows as they change model routes.

Short line:

One app for the work. Your choice of AI.

Supporting line:

Bring the AI access you already have. CoWork OS is free; provider eligibility, limits, and usage charges remain with each provider.

Message Architecture

LayerApproved messageWhat it explains
CategoryFree, open-source AI super app / everything appWhat CoWork OS is
User benefitOne workspace for real work across code, browser, inbox, files, artifacts, and automationWhy it matters
DifferentiatorOpen, multi-provider, and MIT-licensedWhy CoWork is structurally different
Technical engineOne agent harness across supported model routesHow model flexibility works
Access promiseSupported subscriptions/accounts, APIs, gateways, cloud credentials, and local modelsWhat users can bring
Continuity promiseKeep tools, skills, memory, approvals, agents, artifacts, and workflows when routes changeThe practical advantage
TrustLocal-first state, visible execution, policies, approvals, and source accessWhy users can rely on it

The relationship should always be explained in this order:

supported accounts · APIs · gateways · cloud credentials · local models
                                  ↓
                   open multi-provider agent harness
                 tools · memory · agents · policy · routing
                                  ↓
                         CoWork OS super app
          code · browser · inbox · docs · data · automation · channels

“Agent harness” is a technical and comparison term. It must support—not replace—the plain-language super-app category on acquisition and onboarding surfaces.

Positioning Pillars

1. One super app for real work

CoWork keeps the task, the tools, and the output together. A user can move from research to code, a browser check, an inbox action, a document, a spreadsheet, a deck, or an automation without rebuilding context in a separate AI product.

2. Your choice of AI

CoWork is not defined by a single model vendor. Users choose from supported access routes, switch per task, compare models, configure routing, or add ordered fallback providers. Model-specific capabilities and commercial terms still apply.

3. One harness across model strengths

The surrounding work environment—tools, skills, memory, policies, approvals, agents, artifacts, and workspace context—belongs to CoWork. Changing a model route should not mean changing the operating system around the work.

4. GUI-first, CLI-capable, headless-ready

The desktop application is the primary visual operator console. The cowork CLI and headless daemons extend the same runtime to terminals, servers, and remote operations.

5. Open control and visible governance

CoWork is MIT-licensed and local-first. Execution remains inspectable through timelines, approvals, policies, sandboxes, and explicit remote-provider boundaries.

Audience

Primary audiences:

  • AI power users who currently split work across several model apps and agent harnesses.
  • Developers who want coding agents without making the coding tool their entire workspace.
  • Operators, founders, and knowledge workers who need files, browser work, communication, automations, and agents in the same environment.
  • Teams that need model flexibility, visible approvals, and a path to local or self-hosted operation.

Competitive Frame

Coding agents

Claude Code, Codex, Cursor, and OpenCode are primarily evaluated as coding environments or coding-agent harnesses. CoWork should not claim automatic coding superiority. Its durable contrast is a broader work surface: code plus inbox, research, browser, office artifacts, channels, automations, agents, and headless operations.

AI workspaces and super apps

ChatGPT and Claude Cowork increasingly span knowledge work, files, connectors, artifacts, automation, browser use, and project work. CoWork competes in that broad category while differentiating through MIT licensing, model-route flexibility, optional local inference, CLI/headless operation, and user-controlled runtime state.

Open personal-agent systems

OpenClaw and Hermes are open, multi-provider peers. CoWork should differentiate through its GUI-first Everything Workbench, artifact editing, inbox and channels, visual approvals, operator surfaces, and the combination of desktop, CLI, and headless modes—not simply by claiming to be open or multi-provider.

Comparison content must be fit-based:

  • “Choose them when …”
  • “Choose CoWork when …”
  • “Use both when …”

Claim Guardrails

TopicApproved wordingDo not claim
Price“CoWork OS is free and MIT-licensed. Provider usage may cost money.”“All AI is free.”
Breadth“Dozens of model routes and compatible endpoints.”“Every LLM” or “all models.”
Accounts“Supported account-based and subscription routes.”“Any subscription.”
Lock-in“Reduce model-provider lock-in.”“No lock-in.”
Locality“Local-first persistence with optional local inference.”“Everything is fully local.”
Credentials“Credentials are stored locally and sent only to the configured provider or gateway for authentication.”“Keys never leave the device.”
Analytics“No mandatory product analytics by default.”“No telemetry.”
Capability“One harness across different model strengths.”“Every model supports identical features.”
Competition“A broader open workspace for users who want model choice.”“Better than every harness.”

Use stable capability language instead of hardcoded provider, channel, skill, connector, or test counts unless the number is generated from a canonical released source.

Provider and Access Proof

docs/providers.md is the public contract for model access. It must distinguish:

  1. Supported account or subscription routes.
  2. Direct API-key routes.
  3. Compatible gateways and cloud credentials.
  4. Local inference.
  5. Delegated external agent runtimes.

Every marketed route needs a current smoke check for setup, model discovery, text response, streaming, tool use where supported, refresh or expiry behavior, logout/disconnect, and clear entitlement/billing errors.

Comparison Evidence Policy

Comparison pages must use official sources and record:

  • Product and version or release channel.
  • Claim being made.
  • Source URL.
  • Verification date.
  • Stable, experimental, beta, or planned status.
  • Review owner and next review date.

Use “alternative” for proprietary or single-vendor products. Use neutral “vs” language for open-source peers such as Codex, OpenCode, OpenClaw, and Hermes.

Acquisition Surfaces

All high-signal surfaces must share the same hierarchy:

  1. Website hero and social preview.
  2. README and npm description.
  3. Documentation homepage and metadata.
  4. First-run onboarding.
  5. AI & Models / Model Access settings.
  6. About & Updates.
  7. Comparison and migration pages.

Historical changelog and release-note entries should remain historical. Apply this positioning to current summaries and the next release entry only.

Success Criteria

The positioning is working when a new user can answer three questions after the first screen:

  1. What is CoWork OS? — A free, open-source AI super app for real work.
  2. Why is it different? — One open harness across supported model routes.
  3. What can I use with it? — Supported accounts, APIs, gateways, cloud credentials, and local models, with provider terms and charges remaining separate.