Business value in one sentence
Connect enterprise systems to AI agents in minutes, with policy control and measurable ROI per tool call.
ArcaX is the Enterprise Agent Control Plane that mediates every agent-to-legacy interaction through five specialized fabrics — Core, Security, Governance, Observability, and Pod Manager — plus a dedicated SOC pod constellation. Stateless MOE router, Kafka backplane partitioned by tenant and scope, KEDA per-pod autoscaling. Governed. Observable. Air-gap ready.
This is what decision-makers need in the first meeting: a clear business promise, measurable outcomes in days, enterprise-grade controls, and one path to launch.
Connect enterprise systems to AI agents in minutes, with policy control and measurable ROI per tool call.
Run 3 production-like use cases quickly: finance, operations, and security-governed automation.
mTLS, OPA, immutable audit trail, and on-premise posture make security and compliance a built-in default.
No cloud account. No YAML sprawl. No SDK. Clone, configure, register your first source — your agents call it via the MCP protocol instantly.
Authenticate with your ArcaX license key against the private registry, then start all services with one command.
# License key required — request at www.arcax.ai/#demo docker login registry.arcax.ai \ --username $ARCAX_LICENSE_KEY curl -sO https://install.arcax.ai/docker-compose.yml cp .env.example .env docker compose up -d # → 5 services up in ~38s # → Gateway: http://localhost:8000 # → Dashboard: http://localhost:3000
One POST call — ArcaX probes the source, builds the UCS, and generates all MCP tools automatically.
curl -X POST localhost:8000/api/v1/sources \ -H "Content-Type: application/json" \ -d '{ "name": "sap-erp", "type": "sap", "config": { "odata_url": "https://s4hana.corp/sap/opu" } }' # → 8 tools generated in 312ms
Every tool is served via the standard MCP protocol. Point any MCP-compatible agent at the gateway URL.
POST /mcp/acme/3f2a8c/sap_listSalesOrders { "intent": "Q1 2026 open orders", "kwargs": { "limit": 50 } } # ← 200 OK | p99: 1.2ms # ← {"results": [...], "count": 43}
MCP path formula
/mcp/{tenant_id}/{source_id}/{tool_name}
Every AI agent project hits the same wall: how to connect tools to enterprise systems securely, reliably, and with measurable value. ArcaX is that connection — standards-based, governed, and observable.
Tools are compiled and cached on first call. L1 in-memory + L2 Redis. Sub-millisecond for hot paths. No restarts needed.
Every data source is registered, versioned, and lifecycle-managed. Survives gateway restarts. MuleSoft-style environment promotion built-in.
Every call is metered in bytes_in + bytes_out per source, tool, and tenant. Daily quotas. Automatic chargeback. No more spreadsheet guessing.
Ed25519 PKI, auto-enrolled ephemeral leaf certs, HSM-backed key storage, mTLS 1.3, OPA policy on every tool invocation.
API Products, SLA Plans, Subscriptions, and License entitlements. The complete monetisation stack for internal or commercial API programs.
OTLP traces, alert policies, immutable audit log, and a real-time dashboard with D3.js nerve map of all active connections.
Your source schema changed. ArcaX already knows. JIT cache invalidated, new tools live in 30 seconds. Zero restart.
5-stage deterministic pipeline: Probe → UCS → Contextualise → MCP Generate → JIT Cache. <50ms P99.
OTLP traces on every tool call. Custom metrics. Alert policies. OPA hot-reload — all without a restart.
Change Data Capture runs continuously across every registered source. SQL schemas, REST APIs, SAP OData metadata — any drift is detected in under 30 seconds, the JIT cache is invalidated, and new tools are regenerated automatically. No restart. No manual intervention. No agent downtime.
ArcaX is the Enterprise Agent Control Plane: a Mixture of Experts fabric where each plane handles one critical concern. Core mediates, Security trusts, Governance controls, Observability sees — and Pod Manager isolates every agent runtime in its own gVisor bunker.
ArcaX governs what the organization authorizes agents to do: every tool call executed, secured, governed and observable — across clouds, models, applications and legacy environments.
Bounded agent sessions with per-session budgets, trajectory limits and an agentic circuit breaker with manual emergency stop.
Governed MCP tools generated from 100+ enterprise sources in minutes, with HAM governance operations and a versioned tool registry.
SAP, mainframe, relational, NoSQL, messaging and SaaS probes — with self-healing mediation that re-maps tools automatically when schemas drift.
mTLS 1.3 with an internal Ed25519 CA, scoped API keys with mandatory expiry, multi-tenant RBAC and SSO via Keycloak.
OPA policy decisions on every call, shadow policy simulation, human-in-the-loop approvals and adaptive behavioral quarantine.
OpenTelemetry traces, immutable WORM audit ledger and Merkle-sealed audit evidence verifiable offline.
Execute — Secure · Governed · Observable.
Enterprise AI teams spend 60–80% of project time on tool integration, not on agent logic. Every API call is an invisible cost. ArcaX turns that invisible cost into a tracked, governed, optimised line item.
Estimate based on industry benchmarks. Contact us for a project-specific assessment.
Each source passes through a deterministic pipeline: discovery, normalisation, contextualisation, MCP code generation, and JIT caching. No hallucination. No guessing. Certified tool schema every time.
▼ Click any stage to see source-specific detail
Introspects source: OpenAPI, SQL schema, SAP CDS, COBOL copybooks
Universal Capability Schema — normalised JSON pivot: operations, typed fields, source ref. Same format for SQL, REST, SAP, Mainframe, Kafka.
Smart field mapping + OPA guard injection per tool
Emits Pydantic v2 models + FastMCP decorators at runtime
Write L1+L2, serve to agent. <1ms on hot path.
ArcaX deploys as a constellation of single-purpose pods — Core, Security, Governance, Observability, Executor, and Pod Manager — communicating over a Kafka backplane partitioned by tenant and scope. KEDA scales each pod independently on its own queue lag, while a versioned shared kernel keeps security and audit schemas consistent across the fleet.
ArcaX goes beyond a gateway. It's a complete MCP programme stack: SLA Plans, MCP Artifacts with tool filters, self-service key issuance, license entitlements, and alert policies — built-in, no plugins.
Bundle your tools into versioned MCP Artifacts. Define a tool_filter, set a quota, choose a lifecycle state: draft → published → deprecated → retired. Consumers subscribe and get a scoped MCP key — no access beyond what you defined.
Subscriptions auto-generate a scoped key on approval. Keys can be suspended, reactivated, or rotated live — zero gateway restart required.
Enterprise licensing with hard limits: max_sources, max_seats, max_rps. A live usage counter tracks consumption. License keys bind to a tenant — not a cloud account.
Five configurable metrics per source product: error_rate_pct, quota_pct, latency_p99_ms, call_count, bytes_kb. Threshold breaches fire to webhook + email. A live firing panel shows active incidents.
Every configuration change, key issuance, and tool call writes to an append-only source_audit_log. Before/after JSON snapshots. Actor identity, IP, and nanosecond timestamp. PCI-DSS and RGPD ready — no DELETE, ever.
Promote a source configuration from DEV → UAT → PROD with a single API call. Config is cloned, not copy-pasted. Credentials are vault-swapped per environment. Full audit record on every promotion.
Every connector — from SAP BAPI to mainframe TN3270 — goes through the same JIT mediation pipeline, the same OPA policy engine, and the same FinOps metering. Zero exceptions.
Security is not a feature layer — it is the pipeline. Every component is built around zero-trust: no implicit trust, no shared secrets, no plaintext in transit.
Self-contained Certificate Authority. AutoEnrollment issues 1-hour ephemeral leaf certs JIT on first connection. Keys never touch disk after boot.
All inter-service and agent connections enforce TLSv1_3 minimum via MTLSContext. Mutual authentication. Certificate pinning per source.
Set HSM_LIB_PATH to your PKCS#11 library. Keys never reside in process memory. Falls back to FileKeystoreAdapter (chmod 600) without HSM.
Every tool call evaluated against Open Policy Agent before execution. Policy bundles in config/policies/. Gate by tenant, product, tool, time.
Every bridged message carries an Ed25519 signature (__sig__). Downstream systems can verify authenticity and guarantee non-repudiation.
Append-only source_audit_log. Before/after JSON snapshots, actor identity, IP, nanosecond timestamp. PCI-DSS by design. No DELETE ever.
Intermediate CA issues node certs; root key stays offline and HSM-backed. Cert chain depth enforced on every mTLS handshake.
Nonce window (300 s default) backed by Redis Sorted Set. Replayed requests rejected with 401 REPLAY_REJECTED before any tool execution.
All sensitive blobs (keystore, audit snapshots) encrypted with AES-256-GCM. Random 96-bit IV per record. Key derived from ARCAX_MASTER_KEY.
Built-in SOC dashboard page: auth-failure KPIs, IOC/IOA hit counters, MITRE ATT&CK tactic distribution, and a live security event timeline.
The built-in WebSocket dashboard runs on port 3000. Real-time D3.js nerve map, connector health pills, tenant and project control plane, project token lifecycle, FinOps analytics, and zero-trust architecture monitoring.
D3.js force-directed graph of all active agent→gateway→source connections in real time.
Health pills, lifecycle states, tool counts, and one-click promote per registered source.
Full lifecycle management: create, version, publish, deprecate, retire — with subscriber count per product.
Pricing card grid. Create tiers, set rpm / quota_kb_day, assign price — and attach to products instantly.
Consumer key issuance, approve / suspend / cancel workflows, and live call quota meter per subscription.
Entitlement management: max_sources, max_seats, max_rps. Live usage counter and expiry tracking per license key.
Active incidents surface in the firing panel. 5 metrics: error_rate_pct, quota_pct, latency_p99_ms, call_count, bytes_kb.
Immutable event stream. Actor, IP, timestamp, before/after state. Filter by source, tenant, or time range. WORM-compatible export.
Bandwidth analytics: bytes_in + bytes_out per source type. Daily quota gauges. Cost breakdown by source, tool, and tenant.
Per-tenant cost allocation. Export chargeback reports to CSV for internal billing. No spreadsheet gymnastics.
Super admin creates tenants, delegates tenant admins, provisions projects, and issues project-scoped tokens from the same command center.
Watch the control plane, network control plane, pod bunker fleet, tunnels, targets, and zero-trust signals in one operational view.
Browse all 73 supported source types. Filter by category. View schema, authentication requirements, and sample queries.
mTLS cert status, OPA policy bundle version, PKI CA health, HSM connection, and Redis cluster status — at a glance.
Auth failures, sig failures, IOC and IOA hits, quarantined agents. D3 bar chart, MITRE ATT&CK distribution, security event timeline. Polled every 10 s.
The HAM MCP server exposes 17 governance operations as callable tools. Any MCP-compatible agent with the ham:admin scope can issue tokens, manage subscriptions, control licenses, and drive the full agent lifecycle — no API docs needed.
auth.issue_token, auth.refresh_token, auth.revoke_token — full JWT lifecycle from any agentic workflow.
auth.issue_token
auth.refresh_token
auth.revoke_token
subscriptions.list/create/suspend/activate/cancel — manage tenant subscriptions programmatically.
subs.create
subs.suspend
subs.cancel
licenses.issue/verify/revoke/usage — issue keys, check validity, revoke, and track consumption.
lic.issue
lic.verify
lic.revoke
agents.enroll/quarantine/activate/decommission — full FSM lifecycle for every registered agent.
agents.enroll
agents.quarantine
agents.decommission
plans.list — enumerate SLA plans with pricing and entitlements.
plans.list
X-Trace-Id forwarded on every tool response. ham:admin scope enforced. Every call auditable.
X-Trace-Id
ham:admin
| Feature | ArcaX (UAM) | Kong / Axway | Apigee | MuleSoft |
|---|---|---|---|---|
| Native MCP Gateway | ✓ Native | ✗ | ✗ | ✗ |
| JIT Tool Compilation | ✓ | ✗ | ✗ | ✗ |
| 100% On-Premise | ✓ | Partial | Cloud | Hybrid |
| Per-call ROI tracking | ✓ Built-in | Plugin | Plugin | Add-on |
| Auto PKI / mTLS | ✓ Zero-config | Manual | GCP-only | Add-on |
| Air-Gap Deployment | ✓ | Partial | ✗ | ✗ |
| Mainframe Connector | ✓ Native | ✗ | ✗ | Add-on |
| Source onboarding time | < 5 min | Days | Days | Weeks |
| API Marketplace built-in | ✓ Full stack | Plugin | ✓ | Add-on |
| Command Center Dashboard | ✓ 12 views | Basic | Basic | Add-on |
| Vector DB Connectors | ✓ 5 native | ✗ | ✗ | ✗ |
| SIRH/HCM Connectors | ✓ 6 native | ✗ | ✗ | Some |
| Capital Markets Connectors | ✓ 12 native | ✗ | ✗ | ✗ |
| Telecom / OSS-BSS Connectors | ✓ 10 native | ✗ | ✗ | ✗ |
| DEV→UAT→PROD Promotion | ✓ One API call | Manual | Manual | Manual |
| Immutable Audit Log | ✓ PCI-DSS | Plugin | Paid | Paid |
Whether your data lives in SAP, Workday, a mainframe, or a data lake — ArcaX gives your agents access in minutes, not months.
An LLM-powered finance agent queries SAP S/4HANA for open invoices, Oracle ERP for cash position, and Kyriba for FX exposures — via a single gateway, with full audit on every call.
A maintenance agent reads TimescaleDB sensor time-series, queries COBOL copybooks on the mainframe for asset master data, and cross-references Infor ERP for spare parts inventory.
Unify employee data from SAP SuccessFactors, Workday, ADP, and Lucca. An HR agent can answer headcount, leave balance, and compensation queries across all HRIS sources simultaneously.
v0.3 turns governance into a daily tool: a Security Center with a global posture score, scoped API keys with mandatory expiration, source security scanning, an agentic circuit breaker with one-click emergency stop, shadow OPA policies, and Merkle-sealed audit — plus a dashboard in 20 languages.
Security Center — posture score, critical alerts, expired keys, OPA denials and suspicious activity on one page
Agentic circuit breaker — per-session call, repeat, duration and token budgets with one-click emergency stop
API key security — read/write/admin/mcp scopes, mandatory expiration, last-used IP, scope-preserving rotation
Source security scanner — scheme whitelist, SSRF and private-address blocking, cleartext-password detection, TLS posture
Shadow governance — new OPA policies evaluated on live traffic without enforcement, with before/after impact reports
Merkle-sealed audit — tamper-evident audit batches with offline verification and end-to-end correlation IDs
Kimi auto-detect — describe a source in plain language; the LLM pre-fills the declaration form
Dashboard in 20 languages — full UI translation, including Arabic RTL
The platform evolves continuously. Features below are confirmed for upcoming releases.
Dynamic composite tools — governed multi-step workflows (DAG) with dry-run and rollback
Digital twin mocking — validate dangerous writes on an ephemeral, masked sandbox twin
Semantic cache — tenant-isolated result reuse with conservative read-only prefetch
WASM transpiler — signed, sandboxed adapters for legacy wire protocols
Adaptive risk scoring — from observe-only mode to step-up approval gates
ANSSI qualification dossier
Get a personalised demo and a project-specific ROI estimate. No commitment. No cloud dependency required.