• PRO
  • /
  • SDK

Enterprise Licensing for Features Across:

  • Control
  • Create
  • Automate
  • Extend

The AI your enterprise runs. Not the AI you rent.

Cephable controls, creates, and automates across every app your team already uses — with inference running entirely on the device. No cloud routing. No data egress. No per-task meter.

Book a briefing
SCROLL TO WATCH IT WORK

WHERE THE WORK ACTUALLY LIVES

Most of your work never needs the cloud.

Day to day, the bulk of what your teams do is high-frequency and low-to-mid complexity. Cephable runs all of it on-device at zero marginal cost. Escalate to a frontier model only when a task genuinely demands it. Hover any bar to see where real work falls.

  • On-device Create
  • On-device Automate
  • Frontier Value Realized
  • Control, Create and Automate run on your device. Frontier is where a task earns the cost of a cloud model — the most ambitious Automate jobs straddle the line.

The cost model

Routine AI work shouldn't run on a meter.

Cloud AI bills every prompt. Cephable runs on the device and is licensed per user per year, with no token or consumption fees. Put in your own numbers to see the AI spend you'd stop paying.

Knowledge workers

The AI assistant subscriptions you pay for today, and how much of that work moves to the device.

In US dollars.

Percent, 0 to 100. Start conservative.

Developers

Cephable doesn't replace cloud coding tools. It takes the scaffolding, documentation, test, and automation work that would otherwise burn their tokens.

In US dollars. Replace with yours.

Percent, 0 to 100.

AI spend you'd stop paying

$270,000

per year, estimated

Assistant subscriptions absorbed
$144,000
Developer tokens moved on-device
$126,000

This is before your Cephable license. The full calculator nets out the license and any hardware, shows payback and a five-year view, and exports a one-page summary for finance.

Build the full model with these numbers

The savings show up in the product, too. Cephable meters the tokens each prompt uses and shows what the same work would have cost on a frontier cloud model, so IT and finance see the avoided spend as it happens.

What it does

One assistant, inside every app your teams already use.

Cephable works in and with the application in front of the user, not in a separate chat window. It runs at the operating-system level, so it follows people across every app on the device instead of one vendor's ecosystem.


  • Control

    Operate any application, and the operating system itself, by voice, dictation, or contextual quick-action buttons. Prebuilt control training, with optional customization, covers hundreds of apps.

    For exampleNavigate Salesforce and dictate call notes without touching the mouse.


  • Create

    In app contextual generation, revision, summarization, and translation right where people are working. Every one runs locally.

    For exampleSummarize a 40-page PDF and rewrite the brief in your own voice. Nothing leaves the device.


  • Automate

    Completes whole multi-step workflows: research, build the files, send the email, run the commands. It works in the background while the user keeps working.

    For example"Analyze this folder of data and pull out the high level results in a specific format, write a report for the team and draft a summary to the executives."


  • Extend

    Call the on-device agent from your own scripts and tools through a local, OpenAI-compatible endpoint. Hand it your own tools or MCP servers for a single run.

    For examplePoint an existing OpenAI-compatible client at the local endpoint, and the work runs on the device instead of the cloud.

  • Every way people workVoice, keyboard, contextual buttons, and AI workflows, all available at once.
  • 8 languages for voiceInterface, voice controls, and dictation, all running offline on the device.
  • 36 languages for AI tasksGenerate, rewrite, summarize, and translate, offline on the device.

Pick your seat

One architecture. A straight answer for every chair.

Each leader asks a different first question. Pick yours.

For the CEO

AI for the whole workforce, without handing the company's data to someone else's cloud.

  • It runs on the fleet you already own.Every device runs Cephable on its CPU today and uses the GPU and NPU where they exist. No fleet refresh is required to start.
  • It meets people where they work.Cephable works inside Salesforce, SAP, browsers, and the rest of the stack, not in a separate chat window people have to remember to open.
  • It isn't tied to one vendor.Cephable isn't a Microsoft or Google feature. It works across every app on the device, at the same time.
See what it runs on

Security and governance

Private by architecture, not by policy.

Cephable runs as an installed app and inherits the roles and governance you already assign to each user. If you govern the laptop, you already govern Cephable.

Never leaves the device

Processed and stored on the endpoint only.

  • AI inferenceEvery Intelligent Task and every Automate workflow runs locally.
  • Prompts, documents, and generated content
  • Camera and microphone streams
  • Run history and learned preferencesEncrypted to both the Cephable account and the operating-system user.

Cephable does not train its models on customer information.

What uses the network

None of it carries AI inference.

  • Sign-in and authentication
  • Application updates
  • Optional profile sync
  • Web research, when a workflow calls for itThrough Cephable's built-in browser, which follows the device's network and endpoint security rules.

Governance controls


  • Security by inheritance

    Cephable acts with the signed-in user's permissions and can't do anything that user couldn't. There's no separate agent-governance layer to build and reconcile.


  • Rules checked before every action

    Block apps and sites for the whole organization, a team, or one person. Rules are enforced at the tool-call level, not in the prompt, so prompt injection can't talk past them.


  • A person approves the big steps

    Cephable asks for confirmation before deleting or moving files, before running terminal commands, and whenever it needs a decision.


  • File access you scope

    Limit Cephable to one folder and its subfolders, or grant the whole machine. The scope covers reading and writing, set by the user or by IT policy.


  • Encrypted, exportable history

    Workflow runs are kept in encrypted local history that only Cephable, running for that user, can open. It exports to JSONL for third-party audit.


  • No screenshots, no screen recording

    Cephable reads apps through the operating system's accessibility interfaces and structured UI tree. It sees background apps too, without capturing an image of the screen.

Compliance

  • ISO 27001 certified
  • SOC 2 certified
  • HIPAA self-attested*

*HIPAA compliance is self-attested, not third-party audited. Read Cephable's HIPAA statement.

Hardware and models

It runs on the fleet you already own.

Cephable uses all the compute a device already has. It sends each job to the processor that suits it, frees that processor when the job is done, and the person keeps working the whole time.

  • CPU

    Every device

    Runs Cephable today on every machine you own. Handles orchestration and lighter work, and takes over heavier work when no GPU memory is free.

  • GPU

    Most current devices

    Carries the long agent workflows. Integrated graphics count, so machines without a discrete GPU still accelerate.

  • NPU

    Modern AI PCs

    Handles light work that runs continuously, like voice processing, efficiently and at low power.

Why the NPU isn't the whole story: an Automate workflow needs a working memory of at least about 10,000 tokens, and the most powerful current NPUs top out at a fixed context of roughly 1,000. So the heavy work runs on the GPU, or on the CPU when no GPU memory is free.

Cephable Gemma, sized to each machine

Cephable picks the model size that fits each device at install. Users can adjust it. Bigger hardware gets a bigger model and longer workflows.

Five sizes, S through XXL. S is about 2 billion parameters and M about 4 billion. The larger sizes are for machines with substantially more memory.

  • Open foundation. A fine-tune of Google's Gemma 4, chosen for its tool-calling strength and released under the Apache 2.0 license.
  • About 64 silicon-optimized builds. Cephable detects what the device can do and selects the right one.
  • Every major silicon vendor. Intel Core Ultra, AMD Ryzen AI, Qualcomm Snapdragon X, and NVIDIA.

What it needs

No AI PC required. Newer hardware simply runs faster.

16 GB
Memory for the full experienceThe minimum for Automate and other full workloads.
8 GB
Memory for Control and CreateMachines with 8 GB run voice control and Intelligent Tasks.
0
Hardware you have to buy firstEvery device runs Cephable on its CPU and uses the GPU and NPU where they exist.

The better the device, the better the assistant.

Deployment

Rolled out like any other endpoint app.

Built for IT. Cephable deploys in hours as an endpoint rollout, not as a weeks-long integration project.

Rollout
Push it with the device management tools you already run, or through the Microsoft Store. Updates arrive automatically.
Identity
Single sign-on with the identity provider you already use.
Administration
An IT Admin Portal with a governance dashboard. Allow or block apps and sites org-wide or per person, and add rules through the API as JSON.
Licensing
License management with seat assignment and reporting, plus a dedicated, named account manager.
Platforms
Windows and Mac. Runs inside virtual desktop infrastructure, faster with a virtual GPU and on the CPU otherwise.
Memory
16 GB for full Automate workloads. 8 GB runs Control and Create.
Languages
8 for the interface, voice controls, and dictation. 36 for Intelligent Tasks.
Offline
Core features keep working with no connection, because there's no cloud round trip to lose.

Keep your frontier model for the work that earns it.

Cephable takes the high-frequency work onto the device. Your cloud AI keeps the jobs that genuinely need it, and your spend shrinks to those.

Book a briefing Run the ROI calculator

Frequently asked questions

Questions about Cephable for enterprise

Security and governance

Cephable inherits the governance you already apply to the device. It runs from the operating system up, with the signed-in user's permissions, so it can't reach any file, app, or site that the user can't. If you already govern the laptop, you already govern Cephable. Admins can then add Cephable-specific rules on top.

No. Cephable enforces admin rules at the action level, not in the prompt. It checks every step against the rules before it acts, so no wording can talk it past a block. Admins can allow or block apps, sites, and actions for the whole organization or per user. Rule changes reach every device in real time and can stop a task mid-run.

No. Cephable admin analytics show which tools were used, when they were used, and how many tokens they consumed. They don't show prompts, files, or outputs. Prompts and results stay encrypted on each device. Security teams can export history locally when policy requires it. Admins can also switch off remote automation and public web research for the whole organization.

Yes. Cephable's built-in browser follows the same network and endpoint security rules as any other browser on the device, and it can't route around them. Prompts are processed locally, so there's no AI traffic leaving the device for your security tools to inspect. Cephable provides its network requirements during deployment planning.

Deployment and IT

IT deploys Cephable through its normal device management tools and connects it to the identity system it already runs. Cephable supports SSO over SAML 2.0 and OAuth 2.0 with Microsoft Entra ID and Okta, plus SCIM provisioning. Admins set organization rules at portal.cephable.com, and those rules sync to each device when the user signs in.

Cephable sends each job to the processor that suits it best. The NPU handles lightweight work that runs continuously and fits its small context window, like voice processing. Long agent tasks need more working memory, so they run on the integrated or dedicated GPU, with the CPU as a fallback. No AI PC is required, but newer hardware runs faster.

Cost and performance

No. Cephable puts almost no load on the machine while it's idle. When it runs a task, it watches the resources that are free and splits work between the CPU and GPU. If graphics memory is busy, it loads the model into system memory instead. At install, it sets up the model optimized for that device's hardware.

Cephable is licensed per user per year, with no token or consumption fees, so your AI cost is set when you buy. Everyday work runs on the device at no added cost. Teams that keep a frontier model can save it for the few jobs that need one, which shrinks cloud spend to the exceptions.

Last updated