The governed agent runtime
for AI work.
AI agents are shipping into production. LumenFlow governs the work they do across software delivery, operations, and customer-facing systems. We lead with software delivery because it is the clearest enterprise wedge, then expand into the broader agent-runtime surface.
AI agents are ready.
Trust infrastructure isn’t.
Foundation models crossed the capability threshold in 2025. Enterprises are deploying AI agents into production workflows — software delivery, operations, data pipelines, and customer-facing systems. But CISOs, compliance teams, and CFOs block deployment without answers to three questions:
LumenFlow helps teams answer those questions on the supported paths where its policy and evidence controls are active.
Kernel, Cloud, and Sidekick.
Kernel
The local enforcement layer. Policy evaluation, evidence capture, and fail-closed action gates for supported execution paths. Built to be governed, inspected, and validated before covered actions execute.
Cloud
The hosted control plane. Approval routing, exportable evidence storage, team dashboards, cost analytics, workspace isolation, and a control-plane SDK for governed external runtimes. Start with the kernel, add Cloud when you need team governance.
Sidekick
The consumer AI operator built on the same governed runtime. Connect email and messaging channels and delegate real tasks. Sidekick checks workspace policy before governed tool dispatches and records evidence for those actions.
Not a pitch. A product.
LumenFlow is live in beta with real users. Here is what’s in production today.
Policy enforcement
Workspace-scoped policies define what agents can do autonomously and what requires approval. Fail-closed by default.
Exportable evidence
Supported governed actions produce structured receipts with actor, action, and timing. Scope and cost fields are included when captured, and records can be exported for audit.
Per-action cost tracking
Supported model usage is attributed to workspaces with provider and cost detail when available, alongside budget controls and alerts.
Managed inference & BYOK
Start with managed inference for zero-config setup, or connect your own model API keys for full provider choice. Keys encrypted at rest with AES-256.
MCP-native connectivity
Model Context Protocol support lets configured MCP tools run through LumenFlow's admitted tool path, subject to workspace policy and server compatibility.
Kernel enforcement layer
Kernel-grade policy evaluation, evidence capture, and fail-closed action gates packaged as the enforcement layer beneath the hosted control plane.
Governance becomes the platform.
AI agents are moving into ongoing work across connected tools. Major AI platforms are building agent experiences with configurable rules, approvals, and execution environments.
LumenFlow’s shipped wedge is specific: Sidekick checks workspace policy before tool dispatch and records the requested, allowed, and enforced scope with evidence receipts. That gives teams a governed workspace and a tamper-evident record for work handled by this runtime.
The company that controls the governance plane for AI agents sits at the most strategic position in the enterprise stack.
Early stage. Real product.
We’re pre-revenue and honest about it. What we do have is a production system that works.
Live in beta
Users run workflows with policy checks, evidence, and cost tracking on supported execution paths.
Organic adoption
Zero paid acquisition. Users find us through word of mouth and try it because the product speaks for itself.
Production infrastructure
AES-256 at rest, TLS 1.3 in transit. Bring-your-own model keys. Workspace isolation. Not a prototype.
Shipping weekly
Trunk-based development with automated quality gates. Production features shipping every week.
The market is active. Our proof must stay specific.
Major AI platforms offer connected-app access, configurable rules, and review and approval, with tool use, observability, and hosted or self-hosted execution. Major cloud providers and enterprise software vendors also ship agent identity, runtime, and governance capabilities, while work-management and GRC vendors address overlapping needs. The market is active; comparisons should stay specific to each product’s shipped scope and evidence.
LumenFlow is its own governed workspace and Sidekick runtime. Supported external-runtime connectors extend that first-party product where customers need them.
Major AI platforms offer app access, configurable rules, review, tool use, observability, and hosted or self-hosted execution. LumenFlow’s verified comparison is narrower: Sidekick checks workspace policy before tool dispatch and records requested, allowed, and enforced scope in evidence receipts. We make no blanket claim about competitors’ policy or evidence internals.
Where customers already use enterprise identity and control systems, LumenFlow’s posture is to integrate where appropriate. Its shipped proof is policy enforcement at Sidekick tool dispatch and a tamper-evident evidence record for that governed work.
Four compounding advantages.
Operational evidence
Evidence and policy decisions from supported governed work help teams review controls and investigate outcomes. Their value depends on coverage and data quality, not usage volume alone.
Position in the action path
On supported governed paths, Sidekick checks workspace policy before tool dispatch and records the requested, allowed, and enforced scope. Teams can apply these controls and evidence practices to the workflows they route through the runtime.
Per-action cost visibility
Tool calls and model invocations are cost-tracked where provider, model, and usage data are available, giving teams visibility into observed spend across supported workflows.
Auditability trust
Trust comes from enforced policy, exportable evidence, and a governance surface enterprises can evaluate through product behavior, audit artifacts, and diligence, not black-box claims.
Usage-based with platform leverage.
Governed inference is pay-as-you-go — prepaid credits plus a flat governance fee on the work you route through LumenFlow, no recurring bill on the tokens themselves. Team and Enterprise plans layer on top for scale: more workspaces, more governed events per day, and longer evidence retention.
Free — $0
1 workspace, 1,000 governed events/day, 7-day evidence retention. Full governance features. Built to earn trust.
Team — $49/month
Up to 5 workspaces, 50,000 events/day, 90-day retention. Metered pay-as-you-go inference and bring-your-own model key.
Enterprise — custom
Unlimited workspaces and events, 365-day retention. Compliance add-on, audit exports, and enterprise governance controls. SSO today; SAML and SCIM shipped and enabled per organisation once you connect an identity provider.
Hard-to-replicate infrastructure.
The governance kernel contains technical decisions that compound over time. Each one takes months to build correctly.
Workspace-scoped policies
ShippedAutonomy policies, spend limits, and approval requirements configured per workspace. Fail-closed enforcement for tools requiring human approval.
Evidence receipts
ShippedSupported governed actions write timestamped receipts. Model, scope, and cost fields appear when the action path captures them.
Per-action cost tracking
ShippedSupported model usage records provider, model, token, and cost detail when available, with workspace spend visibility.
Workspace isolation (RLS)
ShippedRow-level security across all tables. Each workspace is a hard trust boundary enforced at the database layer.
MCP-native connectivity
Built on an open standardConfigured MCP tools can run through LumenFlow's admitted tool path, subject to workspace policy and server compatibility.
Kernel-first architecture
ShippedThe governance kernel is a dedicated enforcement layer for policy evaluation, evidence capture, and fail-closed action gates beneath the hosted control plane.
Interested in the control plane for AI?
We’re raising our seed round. Request our investor deck or schedule a conversation.