Aegis
The control plane for enterprise AI

Bring every enterprise Agent inside governance.

Aegis is an open-core (Apache 2.0) enterprise AI governance platform: the data plane (enterprise proxy) runs fully on-premise — model keys, prompts, and tool parameters never leave your network — while the SaaS control plane is optional, adding unified management, policy sync, billing, and audit aggregation. Community edition is free; a License Key unlocks Pro/Enterprise.

Start free with CommunitySee the policy engine
  • Open-source data plane · Apache 2.0, free Community edition
  • Data planes run fully on-premise; sensitive data never leaves
  • Optional SaaS control plane: policy sync / billing / audit
  • Priced per data plane — never per agent, no token markup

Four value propositions

See every Agent

Govern before you scale. Help enterprises build an Agent inventory that captures registration status, ownership, activity, call relationships, and potential shadow Agents, turning invisible AI into managed assets.

Control every invocation

Identity, permissions, and cost must bind to the Agent. Give every Agent a distinct identity and short-lived credentials; enforce least privilege across models, tools, MCP servers, and Skills; and attribute calls, anomalies, and cost to the right Agent.

Protect the data boundary

Keep business data inside the customer environment while extending governance to the cloud. Use a default data-residency evidence-sync model so the control plane receives only the information required for safe operation—not the business data itself.

Measure AI maturity

Governance is not a one-time project; it is a continuous operating capability. Derive maturity scores, trends, and improvement recommendations from Agent registration, public AI control, audit coverage, and permission consolidation.

Platform capabilities

01

Govern every enterprise Agent

Agent inventory, lifecycle, ownership, versioning, and risk identification.

02

Hosted models and credentials

Centralized external model access and API credentials without exposing raw secrets to applications.

03

Govern MCP and Skills

Unified registration, versioning, schema, authentication, and authorization.

04

Audit every invocation

Correlate Agents, models, tools, parameter summaries, result summaries, and cost.

05

Bill and allocate cost

Allocate cost by Agent, team, project, application, and cost center.

06

Evaluate public AI control

Maintain a public AI catalog and measure reachability and control rate through the enterprise proxy.

07

Measure AI maturity

Use registration, control, audit coverage, and governance trends to drive continuous improvement.

Policy Governance Engine

Every invocation is adjudicated in real time — identity, attributes, content, and budget under one engine.

Expression policies (ABAC)

Allow rules written as attribute expressions over model / agent / tenant / usage variables, evaluated on every call; default-deny, explicit allow, DENY wins.

model.startsWith('gpt') &&
  usage.inputTokens < 100000

AI decision evaluation

A judgment model renders a structured verdict before each call — decision / categories / confidence / reason; block can deny, review only tags without false positives.

{
  "decision": "block",
  "categories": ["credential_leak"],
  "confidence": 0.93,
  "reason": "possible hard-coded secret"
}

Content safety checks

Scan the text of every request's messages for PII and sensitive content — phone, ID card, bank card, IP, built-in keywords and custom regexes; block or audit-only.

checks: phone · idCard ·
  bankCard · ipAddress ·
  keywords · customRegex

Rate limits & budget circuit breaker

Per-key requests-per-minute limits and a tenant daily token budget; exceed it with a hard 429 or a one-time alert, with an automatic warning at 80%.

rpmPerKey: 120
dailyTokenBudget: 2000000
budgetHardStop: true
Read the full documentation

Endpoints, the complete expression reference, judgment billing and troubleshooting live in the docs.

Documentation

Platform architecture

Open-source data plane + optional control plane (Community needs no control plane)

Control plane (SaaS or fully private)

Directory, policy distribution, multi-data-plane dashboards, billing, and audit aggregation; fully private in the Enterprise edition.

Enterprise proxy (data plane)

Standalone gateway, locally encrypted keys, MCP/Skill repos, offline policy enforcement, and usage metering.

Prompts, files, tool parameters, and business logs stay inside the customer environment; data planes keep running when the control plane is down.

Three deployment modes

One policy, credential and metering fabric — three deployment shapes you can mix and scale with, from a single seat to enterprise groups, with no re-architecture.

Mode 1 · Zero infrastructure

Desktop proxy gateway

SMBs · teams without a stable VPN

Anyone who needs it runs an enterprise proxy on their own machine, and user-side agents hit 127.0.0.1:8931 locally: no auth, no server, no ops — model keys and prompts never even cross the LAN. Community standalone mode works out of the box with no activation.

  • Download the proxy and pick Community/standalone at first launch — no activation needed
  • Intercepts user-side agents like Claude Code and WorkBuddy locally; policies keep working offline
  • Scaling out = one more proxy per employee, unbounded
Mode 2 · Centralized access

Enterprise central gateway

Mid-size and large enterprises · teams with IT

Proxies run on servers listening on a LAN address; the whole company shares one endpoint (IP:port directory in the console, frontable by an LB/VIP). Remote callers present an agent key, validated offline by the proxy.

  • Stateless, peer proxies — natural load balancing and failover
  • Policies, key rotations and governance prompts propagate to every proxy within 60 seconds
  • Metering is idempotent by traceId — never double-charged across proxies
Mode 3 · Zero deployment

SaaS relay gateway

Teams without ops capacity · rapid evaluation

Nothing to install: agents call the cloud-hosted gateway endpoint directly. The control plane hosts credentials (AES-256-GCM, tenant-isolated), injects them into upstream requests, and enforces policy, content safety and billing in the cloud. Accept the hosted-key risk and go live in minutes.

  • No install, no ops — point one endpoint and start
  • Credentials encrypted at rest, isolated per tenant by API key
  • Same policies and billing as self-hosted proxies — migrate to Mode 1/2 anytime
Three modes, one governance fabric
  • Uniform policy enforcement (default deny, DENY wins)
  • Uniform credential custody (hash-only distribution, keys never leave the control plane)
  • Uniform governance prompt injection on every call
  • Uniform audit hash chain and cost attribution

The Agent governance loop

  1. 1

    Register

    Agent identity, owner, version, and data classification

  2. 2

    Authorize

    Role- and attribute-based policies for models, tools, and quotas

  3. 3

    Invoke

    Call models through one endpoint; keys stay hosted

  4. 4

    Audit

    Immutable event chain: who invoked what, when

  5. 5

    Optimize

    Cost attribution and maturity scores drive improvement

Public AI control evaluation

Reachable is not controlled. The enterprise proxy continuously probes catalog endpoints and quantifies governance coverage.

Reachability score

The enterprise proxy continuously probes catalog endpoints from the workstation and detects whether network policies already restrict them.

Control rate

The share of reachable endpoints covered by allow, block, or approval policies—governance coverage under your enterprise AI policy.

Governance maturity

A continuous score and trend, weighted across asset registration, public AI control, audit coverage, and permission consolidation.

Network blocking and full traffic logging belong to the firewall layer; Aegis provides the catalog, detection, and scoring.

Launch scenarios

Low-sensitivity, high-value: knowledge assistants and automation

Internal knowledge Q&A, document retrieval, ticket assistance, engineering knowledge search, and operations automation. Establish an Agent inventory, enforce least privilege, and attribute cost by Agent and team.

High-sensitivity: engineering and data collaboration Agents

Code assistance, internal data queries, knowledge retrieval, and restricted tool calls. Sensitive data and tool parameters stay inside the customer environment; short-lived credentials restrict invocation scope and create audit evidence.

Plans & pricing

Open-core + hybrid deployment: the data plane (gateway / enterprise proxy) runs in your environment, the control plane is SaaS; enterprise features unlock via license. Upgrade from Community seamlessly.

Community

Free

Self-hosted proxy · Apache 2.0

Open-source data plane core: standalone mode runs fully on-premise, no license, no SaaS. For individuals and small teams.

  • Unlimited data planes (self-hosted)
  • Local key / MCP / Skill repos
  • Local agent token issuing
  • Local usage billing
  • Community support
Start for free

Pro

¥299/month

Self-hosted data planes + SaaS control plane

For 10-50 person AI-native teams: unified policy sync, multi-data-plane dashboards, audit export.

  • Up to 3 data planes
  • 200 agents / 50 members
  • Signed policy snapshots
  • 30-day audit retention + export
  • Email support
Start trial

Enterprise

¥2999/month

Fully private control plane + License Key

For compliance-driven enterprises: data never leaves your boundary, control plane can be fully private, SLA.

  • Unlimited data planes
  • Control plane: SaaS or fully private
  • SSO · advanced RBAC
  • Unlimited audit retention, compliance formats
  • SOC 2 · 4h SLA
Contact sales

Hosted

¥999/month

Fully managed SaaS (data plane on vendor cloud)

For teams without ops capacity: fully managed, ready to use.

  • 50 agents / 20 members
  • 90-day audit retention
  • Zero ops
  • Audit export
Start trial

Billing principle: priced per data plane (Pro ¥299/mo or ¥2,990/yr; Enterprise ¥2,999/mo or ¥29,990/yr) — never per agent, no token markup. Community features stay free forever and upgrade with zero migration.

Who benefits

For CIO / IT

Unified directory, standard identity integration, MCP/Skill asset governance, and operational visibility. Maturity scores bring AI governance to the executive dashboard.

For CISO / Security

Reduce credential exposure, shadow AI, unauthorized tool calls, and data egress risk—turned into verifiable policies, audit events, and continuous scoring.

For AI platform teams

Reuse authentication, credentials, policies, approvals, and MCP infrastructure to launch securely faster—no duplicated build-out per business line.

Governance principles and security boundaries

  1. 01

    Data stays by default: prompts, files, tool parameters, and business logs remain in the customer environment

  2. 02

    Least privilege: no long-lived, shared, or unlimited static keys

  3. 03

    Accountability: key invocations link actor, authorization source, target resource, and cost attribution

  4. 04

    Control belongs to the customer: assets, policies, quotas, retention, and sync boundaries

  5. 05

    Clear boundaries: Aegis does not replace firewalls for network blocking or full traffic logging

Download the enterprise proxy (data plane)

One proxy binary, two shapes: Community standalone mode runs fully on-premise (free, no activation, Apache 2.0); connected mode pairs with the SaaS control plane for policy sync, multi-data-plane dashboards, and billing aggregation.

Aegis enterprise proxy · Community, Apache 2.0
Current version 0.1.0

Release notes: 首个公开预览版：Agent 治理控制面、模型网关、MCP / Skill 权限管理与审计闭环。

Download enterprise proxy v0.1.0
  1. 1
    Download & install

    Download and install the enterprise proxy from this page

  2. 2
    Choose a mode

    Pick Community/standalone at first launch (no activation) or connected mode with an activation string

  3. 3
    Start using

    Configure model keys, MCPs, and Skills in the Local repo, then call through the local gateway

Request an Aegis platform demo

Provide basic information, and Aegis will arrange a tailored demo based on the number of Agents, model usage, and governance priorities.

预约演示

预约演示请联系 aegis@futurelabs.vip，我们将在 1-2 个工作日内回复。

The control plane for enterprise AI | Aegis