Product · Roadmap
What we are building
Our work is grouped into three waves. P0 is the day-one trust surface that gets us past procurement. P1 is the operational depth that makes the product feel finished. P2 is the long-tail differentiators that turn a competitor into a category. Items move between sections as priorities shift.
See the changelog for shipping receipts.
Day-one trust surface
Everything an enterprise reviewer needs to sign off before a single seat is bought: data residency, public security posture, signed agent manifests, and four-eye admin actions.
Data residency in US, EU, and UK
ShippedPick a region at Workspace creation and your data stays there. Inference, email, and retrieval all stay in-region.
Signed agent manifests
ShippedEvery agent is signed at upload and verified at install; tampered manifests are refused at the gate.
Four-eye admin actions
ShippedPrivileged operations require a second admin to approve; both reviewers are recorded in the audit chain.
Hash-pinned approvals on every external write
ShippedAn approver sees the exact payload an agent will execute. The platform refuses to run a payload whose hash does not match.
Public DPA, subprocessor list, and live status page
ShippedTrust signals live at /legal/dpa, /legal/subprocessors, and /status; everything updates automatically when we add a vendor or open an incident.
Cookie consent banner with UK GDPR essential-only default
ShippedThe banner asks once and remembers your choice for six months. Non-essential cookies stay off until you opt in.
Operational depth
What makes the product feel finished: durable runs, per-stage model routing, queue-backed external writes, rich audit exports, and a marketplace that is actually a marketplace.
Durable, crash-safe runs with isolated sandboxes
ShippedAgents pick up where they left off after a transient failure. A sandbox isolates code execution from the platform.
Outbox + queue infrastructure
ShippedWrites to external systems are queued and retried so a connector hiccup never silently drops work.
Per-stage model routing
ShippedEach agent stage picks the model best suited for the task; cheap models do triage, capable models do reasoning.
Marketplace install metering and ratings
ShippedPublic install counts and aggregate ratings backed by signed receipts; reviews require a verified install.
Marketplace payouts via PayFac
In progressKYC/KYB onboarding for sellers, with split card acceptance and payouts. Specific vendors are listed in the subprocessor page.
Per-Workspace audit-chain export
In progressOne-click export of the entire tamper-evident audit chain with the chain head pinned at export time.
Healthcare BAA pack
PlannedBAA-ready Workspace template with PHI-aware connectors, audit retention policies, and de-identification helpers.
Long-tail differentiators
Where the product stops looking like a competitor and starts looking like a category. Customer-tunable safety policies, evaluation runs as a first-class artifact, and marketplace integrations.
Customer-tunable approval policies
In progressPer-tool, per-agent approval thresholds with optional dual control. Configurable in the dashboard.
Evaluation runs as a first-class artifact
PlannedRun an agent against a graded test set and ship the eval report alongside the agent in the marketplace.
Bring-your-own-key model providers
PlannedEnterprise customers can plug their own model accounts in for billing and data-handling separation.
One-click provisioning from a partner marketplace
PlannedSpin up an AgentHub Workspace from a partner marketplace with auto-injected configuration so the first run works without setup.
Public roadmap voting
PlannedCustomers can upvote roadmap items so we know which ones to pull in.
Got a request?
Tell us what you would like next. Email hello@mvsagents.ai with your use case and we will get back to you with an honest answer about timing.