Roadmap
Roadmap
Maia Cloud will run Maia OS for organizations that would rather not operate it themselves. This page lists what comes next, in order. Dates are targets, not commitments, and every Maia Cloud milestone is planned.
Targets
What we aim for next. A target is a goal we are working towards, not a promise.
January 2027
Maia OS ready on canary
The Maia OS capabilities that the private alpha needs are ready to run from a canary build.
plannedJanuary 2027
Maia Cloud private alpha (C3) with a first private-alpha tenant
Accounts and organizations, then the console for the first private-alpha tenant, who runs managed Nodes.
plannedFebruary to March 2027
Design partners
We recruit design partners for Maia Cloud, starting from the requests for the private alpha.
plannedNo date
Public launch
General availability waits for its evidence gates.
planned
Maia Cloud milestones
Each milestone depends on the one before it. Maia Cloud has no tenant feature yet, so every milestone reads planned.
Hosting (C2)
The infrastructure that Maia Cloud will run on: tenant machines and sandboxes, EU regions first, backups, monitoring and incident response. Its owner builds and operates it apart from the console, and Maia Cloud connects to it through an agreed contract.
Depends on: The hosting infrastructure and its owner.
Private alpha (C3)
Accounts and organizations, then the console for the first private-alpha tenant, who runs managed Nodes. It covers tenant provisioning and decommissioning, the first-owner bootstrap, compute choice in the console, export and restore, restore drills, and support access only with the customer's approval, and the first always-on agent computers (see Agent computers and model sign-in below).
Depends on: C2, and the Maia releases it relies on.
Commercial and payments (C4)
Plans, seats, billing and payments, and the legal texts that go with them: the terms, the privacy policy, the processor agreement and the subprocessor list.
Depends on: C3.
Public launch (GA)
Public launch happens when its evidence gates pass. No date is set.
Depends on: C4, and the gates below.
Gates to the public launch
The public launch follows these gates, in this order. Each one needs evidence before the next begins.
- Design-partner letters of intent
- A self-hosted licence
- A managed pilot
- Invited paying customers
- General availability
Agent computers and model sign-in
Two capabilities shape the private alpha and the design-partner phase. Both are planned; neither exists yet.
An always-on computer per agent
Each agent gets its own computer that keeps its files and state. It sleeps when idle, which aims to cut idle compute cost, and wakes on a message or a schedule. A person from the tenant can watch it live, take over, then hand it back. Each takeover leaves an audit entry (who, when, which agent) in the control plane; what was on screen stays in the tenant's environment. The first version, with C3, keeps state, stops when idle, starts on a message and allows live takeover. Waking on a schedule and resuming without a cold start follow in the design-partner phase.
Depends on: C2 tenant machines with a start, stop and wake contract; Maia OS contained execution of vendor harnesses, then routines and event triggers for schedules.
Bring your own subscription
Each tenant signs its agents in with its own Claude or ChatGPT plan instead of an API key. The sign-in stays in that tenant's environment, and Maia Cloud never resells or pools model usage.
Depends on: The always-on computer above, per-tenant secret storage, and each provider's terms for using its plans this way.
Maia OS delivery plan
Maia OS is the software Maia Cloud will run, and it is in private alpha. These phases overlap: a piece starts once what it depends on holds. Every item shows its status word.
Maia ships canary builds, and no stable release is planned.
Finish the foundation
- Acceptance closure and source organizationplanned
- Measurement of the work Maia doessource-implemented candidate
- Portfolio reads and persistent Workspace conversationsplanned
- Open coordination work between Sessionsplanned
Enterprise foundation
- Vulnerability disclosure and supply-chain controlsplanned
- Grants per Workspace and Projectsource-implemented candidate
- Contained execution of vendor harnessesplanned
- Usable local executionplanned
Workflows
- Agentsplanned
- Routines and event triggersplanned
- Tasks and goalsplanned
Evaluation and learning
- Reusable evaluationplanned
- Scorers and routersplanned
- Datasets and custom modelsplanned
Execution at scale and operations
- Server setupplanned
- Execution Nodes and a pool of local and remote workersplanned
- Session capsplanned
Human layer
- Web clientplanned
- Personal agentplanned
Distribution and release
- Signed releases with provenanceplanned
- Seamless updatesplanned
- Windows CLI and SDKplanned
- Self-managed deployment recipesplanned
Later
- Messaging between organizations, with no shared authorityplanned
- More chat connectorsplanned
- Governed browser and computer useplanned
- Short-lived Execution Nodes from an operator-run provisionerplanned
We recruit design partners from the requests for the private alpha.
Request access