Skip to content
Skyfixby

Every model, tool and agent, through one sovereign control plane

One platform for AI gateway (LLM · MCP · A2A), FinOps, observability, governance and security, deployed in your environment under your keys.

Book a demo See the console
The step waterfall of one request trace in the Skyfix console: authentication, policy checks, rate limit, routing, input guardrail, budget check, response cache, the model call, response read and output guardrail, each with its timing.
One request traced through authentication, policy, routing, guardrails, budget and the model call.

One system

One platform, not a stack of point tools

A single Rust control plane and a single gateway carry every pillar. Nothing is bolted on, so policy, cost, traces and security are read from the same request record.

A bolted-on stack

Five tools, five keys, five logs, five ideas of policy, and agent-to-agent traffic that none of them sees.

Skyfix
LLM · MCP · A2A

  1. Identity

  2. Policy

  3. AI Firewall

  4. Budget

  5. Route

  6. Evidence

One ledger

Every decision, hash-chained, signed checkpoints.

One console

One role model, one step-up classification.

  • one policy model
  • one identity
  • one evidence record
  • one console

Enforced by the gateway against a signed, locally verified policy bundle

A chat completion, an MCP tool call, an A2A task, a skill step or a batch line: each one takes the same six stages in the same order and lands in the same record.

Platform

Five pillars. One platform.

Built as one system from the start, not assembled from point tools. Every pillar shares the same control plane, policy model, evidence record and console.

Guided tour

See the console

Eight frames of the enterprise console. Hover or tap a numbered box to see what it shows.

Set the routing policy

A task rule tells Garuda which model tier fits a task, which capabilities to prefer or avoid, and how confident it must be before the rule applies.

The Models › Routing page of the Skyfix console with the Code generation task rule open: its model tier, preferred and avoided capability tags, the confidence it needs and whether it may route.The Models › Routing page of the Skyfix console with the Code generation task rule open: its model tier, preferred and avoided capability tags, the confidence it needs and whether it may route.

  1. Code generation is one of the task rules Garuda recognises; this one was edited and is allowed to route.

  2. The Frontier tier ranks the most capable model in the catalog first and cost second.

  3. Capability tags refine the ranking inside the tier: models tagged for code and reasoning rank higher, lightweight models lower.

  4. The rule applies only when Garuda is at least 85% confident; below that the request stays on default routing.

Sovereign and private

Deployed in your environment

Enterprise edition, delivered into the cluster you run. No request depends on Arka AI Labs infrastructure.

Deployment overview

  • Your Kubernetes

    Installed with a Helm chart into your cluster, with your PostgreSQL, cache and secrets store.

  • Your keys

    Every trust anchor lives in your own secrets store. No vendor key is built into the images.

  • Your boundary

    Requests are authorized inside your installation. Data, evidence and audit stay in storage you operate.

Evidence

What every decision records

LLM, MCP and A2A decisions land in one hash-chained ledger with signed checkpoints. Exports verify offline from your installation's own root fingerprint.

AI audit trail
  • Caller and end user
  • Workspace and key
  • Alias and model or tool used
  • Policy result
  • AI Firewall outcome
  • Budget status
  • Route reason and fallback
  • Provider or server
  • Tokens, cost and latency
  • Trace ID

Model and tool coverage

Works with the models, tools and agents you choose

One standard chat-completions API and other widely used model API formats, plus native MCP and A2A for tools and agents. Bring-your-own-key accounts and private model pools share the same policy and evidence.

Open standards supported by name
Public model providersLLM API
Connected over HTTPS with failover and circuit breakers, behind model aliases your applications call.Unified LLM API
Private and self-hosted modelsLLM API
Cloud-hosted model services and private endpoints in your network, through the same governed pipeline.AI Gateway
MCP tool serversMCP
A registry with per-tool approvals, argument policies, call limits and OAuth-based client authorization.MCP Gateway
A2A agentsA2A
Publish your Agent Card, accept governed inbound tasks and federate to external agents over JSON-RPC, gRPC and HTTP.A2A Gateway

FAQ

Questions buyers ask first

All questions

What is an AI control plane?

One system that governs every model, tool and agent call an organization makes: who may call what, under which policy, within which budget, routed where, inspected how and recorded where. Skyfix is that system, with the AI gateway, FinOps, observability, governance and security as one platform rather than five tools.

How is an AI control plane different from an AI gateway?

The gateway is the enforcement point in the request path. The control plane is everything that decides what the gateway enforces and proves what it did: signed policy, budgets, identity, the evidence ledger, assessments and the console. Skyfix ships both, built as one system.

Where does Skyfix run and who holds the keys?

In your own Kubernetes cluster, installed with Helm, using your PostgreSQL, cache and secrets store. Every signing key lives in your secrets store under one installation root; no vendor key is built into the images, and no request depends on Arka AI Labs infrastructure.

Which APIs and protocols does Skyfix support?

An industry-standard chat-completions API and other widely used model API formats, embeddings, rerank, moderation, image, audio, realtime and batches; the Model Context Protocol for tools with OAuth 2.1; the Agent2Agent protocol for agents over JSON-RPC, gRPC and HTTP+JSON; Agent Skills; OpenTelemetry for export; OIDC, SAML, SCIM and SPIFFE for identity.

Does Skyfix certify compliance with AI frameworks?

No. Framework assessments support evidence for frameworks such as the NIST AI RMF, ISO/IEC 42001 and the EU AI Act by linking controls to recorded evidence. They are not certifications, attestations or compliance scores.

Demo request

Book a demo

Tell us where your teams run AI today and what has to be governed first: model traffic, MCP tools, A2A agents, spend, evidence or identity. A solutions engineer replies to schedule a walkthrough.

Architecture walkthrough for your environment

Control plane, gateway fleet, data stores and trust anchors mapped onto your cluster and network.

Pilot scoping in your Kubernetes cluster

Helm install, identity, a first set of policies and budgets, and the traffic to bring under control first.

Security and procurement questionnaire support

Architecture, data-handling and evidence answers from the platform documentation, not a marketing deck.

Tell us what you need to prove

Share the essentials and we will follow up to schedule a focused conversation about your AI evidence, policy, and audit needs.