Atlas — a thinking partner for strategy.
.wm DSL practitioners already write.For agents and the CLI-curious — the same engine, headless. Published at launch.
Atlas is a strategy-design app for senior practitioners. You write Wardley maps in the
.wm format, open them on a Mac-native canvas, and work through premortems,
critiques, and composed memos without leaving the app. The AI side defaults to challenge,
not validation — every finding carries its confidence, provenance, and review state.
Local-first. No sign-in. Your maps stay on your machine.
Each mode stands on its own and sharpens the others. Everything below is in the shipping build — nothing here is a roadmap promise.
Map the landscape.
A Mac-native canvas over the .wm DSL you already write. Zoom, pan, fullscreen. No browser tab, nothing to sign into.
A thinking partner that defaults to challenge, not validation.
Every finding carries confidence, provenance, and review state. Weak moves surface first; agreement is what remains after disagreement runs out.
ShippedSee the moves that fail before you make them.
A guided pass through failure modes: likelihood, severity, mitigation, each tied back to a component on the map. Exports as a Systema-compatible manifest fragment.
ShippedWhat if you'd assumed the opposite?
Atlas finds the load-bearing assumption your map leans on, inverts it, and regenerates the map under the opposite premise. Origin and counterfactual sit side by side, with every move between them flagged. Branches save; switching never overwrites the original.
ShippedOne project, multiple audiences.
Generate a board memo, an all-hands update, a Slack post, and a one-pager from the same Atlas project. Every output cites the components it came from.
ShippedEvery AI output carries three pieces of metadata: how confident the model is, where the claim came from, and where it stands in human review. The critique panel leads with the model's lowest-confidence findings, not its loudest. Atlas surfaces “I don't know” rather than fill the gap with confident prose.
The default AI runs on-device — Apple FoundationModels where Apple Intelligence is available, otherwise a local open-weight model. Cloud AI is optional and uses your own API key, sent directly from your Mac to the provider you chose. labsly never sees that content. There is no analytics, no telemetry, no account, no usage logging. Your strategy work stays on your machine unless you explicitly send it somewhere.
Every Atlas verb works identically from the app, the CLI, and the MCP server. The Wardley canvas, critique, premortem, counterfactual, and composer all behave the same whether you operate them by hand or whether an agent does. The MCP package — labsly-atlas-mcp — opens this to any MCP client.
One-time purchase. Price announced at launch.
Atlas is a Mac-native strategy workbench. The default AI runs on-device — Apple FoundationModels where Apple Intelligence is available, otherwise a local open-weight model; for more ambitious models you bring your own Anthropic or OpenAI key. Buy Atlas once and own it forever. No subscription. No per-token fee. We don't pay for your tokens, and we don't bill you for them either.
Major versions (every 18–24 months) are a separate one-time upgrade. Your v1 licence keeps working on v1 forever.
Atlas's MCP server is the same engine the app ships. Install it independently if you want Atlas's strategy verbs available to an agent — Claude Desktop, the Anthropic API, any FastMCP-compatible client. The package is built and versioned; it goes to PyPI at launch, so the command below does not resolve yet.
$ pipx install labsly-atlas-mcp
Twelve tools, five verb groups (Wardley, critique, premortem, counterfactual, composer). The same shared response envelope every Atlas verb returns. The Atlas CLI is the underlying engine — the MCP package is a thin protocol shim, not a re-implementation. See the package README for Claude Desktop config.
The motion a screenshot can't carry — map, critique, premortem, counterfactual, composer, end to end.
| You are doing this with | That gets you | What Atlas changes |
|---|---|---|
| OnlineWardleyMaps (browser, community-funded) | The standard .wm format, a browser canvas, and a working community. Not a native app; no critique, no premortem, no composer. |
Reads the same .wm format so your maps port over. Adds a Mac-native canvas, critique that defaults to challenge, premortem, counterfactual, and composer. |
| Strategy decks (Keynote / Google Slides) | A readable artifact. Not a thinking surface. Decks don't let you move a component on an axis and watch the conversation shift. | A live canvas while you think, and a composed memo after — generated from the map, not copy-pasted into the deck. |
| Tool-stack PMs (Notion + Miro + ChatGPT subscription) | Three apps, two sign-ins, an AI that defaults to agreement, and three monthly bills. Coordination cost eats the practice. | One project folder on your Mac. One AI surface that dissents before it agrees. One purchase, perpetual. |
You own a bet that will outlast your next two jobs. You need a surface that lets you think about it in one piece — the components, the evolution stages, the places your roadmap disagrees with the landscape. Atlas gives you the canvas and then argues with what's on it.
You walk into a room with a client's ambiguity and walk out with a shared map of the moves available. Atlas is the tool that lets you build that map alongside the client instead of presenting it back as a deck. Export is a memo they can circulate internally — not a PDF of your deck with their logo swapped.
You are answering the same strategy question for the board, the team, and the customer. Atlas composes the same project into the three artifacts each audience reads, cited back to the components they came from. One source of truth; three voices.
You already write .wm. Atlas reads it. What's new is the canvas (Mac-native, zoom-pan-fullscreen — shipped), the critique (defaults to challenge — shipped), the counterfactual (what-if-we-inverted-the-load-bearing-assumption — shipped), and the composer (map-to-memo in one step — shipped). The format you know; the surfaces the community has been missing.
A one-time purchase — the price is announced at launch. No subscription, no per-token fee, no auto-renewal. Major versions (every 18–24 months) are a separate one-time upgrade; your v1 licence keeps working on v1 forever. We can sell perpetual because we don't pay for your AI inference — the local default is free, and the cloud path uses your own API key sent directly to the provider you chose.
Atlas Teams arrives post-v1 with shared workspaces and SSO, priced per seat. Until then, every team member buys their own Pro licence.
No. The default AI runs on-device — Apple FoundationModels where Apple Intelligence is available, otherwise a local open-weight model installed on the Mac. If you opt into a cloud provider (Anthropic, OpenAI) Atlas sends your prompt directly from your Mac to the provider using your API key — labsly never sees the content. You can use Atlas entirely offline once the local model path you chose is available; the cloud path is opt-in.
None. No analytics, no telemetry, no crash reporting, no account, no advertising. App Tracking Transparency does not apply — there is nothing to track. The full policy is at /atlas/privacy. The App Store privacy label declares zero data collection across every Apple data category.
No. Atlas is proprietary — labsly, all rights reserved. We're a single-developer commercial product, and the perpetual-licence pricing model depends on the source not being free to fork and resell. The Wardley Mapping technique is open (Simon Wardley, CC BY-SA 4.0) and Atlas attributes it correctly; see /atlas/THIRD_PARTY for the full attribution and Atlas's ShareAlike reading. The MCP server package is also proprietary, distributed as a thin protocol shim around the closed-source Atlas CLI.
No, deliberately. Wardley's canonical published maps — the Tea Shop, GCHQ, HS2, public-health maps — are CC BY-SA-licensed expression, and a proprietary commercial product reproducing them would create ShareAlike obligations Atlas does not honour. The AI is system-prompted to redirect canonical-map requests to a labsly-original equivalent (Pottery Studio, Indie Newsletter, Solo Consultant — all bundled). The technique itself is taught freely.
Yes. The Wardley canvas, the .wm parser, the on-device AI, the premortem flow, the composer, and the counterfactual generator all run offline. The only feature that requires the internet is the optional cloud-LLM path (Anthropic, OpenAI) and only when you've configured a key.
At launch, pipx install labsly-atlas-mcp exposes Atlas's twelve verbs as MCP tools any client can call (Claude Desktop, the Anthropic API, FastMCP-compatible runners). The Atlas CLI — bundled inside the macOS app — does the same work from a terminal. Every verb produces an identical structured response envelope across the app, the CLI, and the MCP server; agents and humans drive the same engine.
Atlas is a one-time purchase on the Mac App Store — no subscription, no per-token fee. The price is announced at launch. The MCP package publishes separately for agents and the CLI-curious, also at launch.