ARC
Pricing
Discord
Talk to an engineerStart free
Start free
ARCby FenxLabsCollective Intelligence

Platform

  • How it works
  • Custom routing and governance
  • Use cases
  • Example architectures

Deployment

  • Hosted Platform
  • Managed Service
  • Self-hosted Licence
  • For ISVs
  • Open the portal

Services

  • Assessment and planning
  • Architecture and integration
  • Model Adaptation
  • EU AI Act readiness
  • Managed Service operations

Resources

  • Cost modeller
  • Compare approaches
  • Pricing
  • FAQ
  • Transparency
  • Get started
  • Discord community

Legal

  • Privacy notice
  • Terms and conditions
  • Sub-processors
  • Accessibility

For individuals

  • FenxChat

FenxLabs. KvK 91762782.

Herengracht 320, 1016 CE Amsterdam, Netherlands

+31 85 060 5273contact@fenxlabs.ai

© 2026 FenxLabs. All rights reserved.

Hosted Platform

Govern AI requests through one hosted endpoint

The Hosted Platform is live today. FenxLabs operates FenxARC™. Connect applications and agents through one OpenAI-compatible endpoint, then enforce which models and tools they may use, which policies apply and what they may spend.

Start freeTalk to an engineer

Whether you are one developer or a platform team, if you are already calling more than one model, ARC is for you. Free tier, no card required. Nothing to deploy.

Change the endpoint, not the application

Keep the OpenAI client your application or agent already uses. A new base URL and ARC key put access, privacy and routing policy into every request.

Three steps to a governed request

  1. 01Create a keyOpen the live developer portal and create a key with the fxk_ prefix.
  2. 02Change the base URLPoint your existing OpenAI client at https://api.askarc.app/api/v1.
  3. 03Choose the request pathOmit the model and ARC scores the request and routes it. Name a model and the request goes there instead of being routed.
Pinning is bounded by the model lists assigned to the key in use.
Start freeSee how a request is handled
quickstart.py
Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.askarc.app/api/v1",
    api_key="fxk_your_api_key_here",
)

messages = [{"role": "user", "content": "Summarize this contract..."}]

# No model named, so ARC applies policy and selects the route.
routed = client.chat.completions.create(
    messages=messages,
)

# Named model, so the request goes there instead.
pinned = client.chat.completions.create(
    model="one-of-your-assigned-models",
    messages=messages,
)
SDK examples cover Python, Node.js, Go, Rust, cURL.

When the Hosted Platform is the wrong fit

Three requirements the Hosted Platform does not meet. If your deployment depends on any of them, read the Self-hosted Licence instead.

  • Request data must not cross your network boundary.
  • The machine learning models you already run must be routing destinations alongside your language models.
  • Your vector storage must sit behind the control plane, so a model reaches only what it has permission to see.
Explore the Self-hosted Licence

Control applications and agents from one place

Give each key access to approved lists of models, tools and rules. Each application or agent request runs inside one list, while administrators change access, privacy and spend policy without redeploying it.

One governed path to approved resources

Applications and agents
ARC Hosted Platform
Approved access to
Approved toolsCalled and returned through ARC
ARC Hosted Platform
Approved toolsCalled and returned through ARC
  • Provider APIs
  • Private endpoints
  • Models you host

Any OpenAI-compatible endpoint becomes a destination. Your URL, your key, your headers.

See the ratings and flags you set

Capability ratings

  • Reasoning
  • Coding
  • Creativity
  • Conversational
  • Speed
  • Web search
  • Enhanced reasoning
  • Image generation

Model flags

  • Vision
  • Streaming
  • Function calling
  • Privacy-compliant
  • Uncensored
  • Task determination
  • Premium

What you control in the portal

Lists and access
A list groups the models, tools and rules approved for a team or use case. Each request runs inside one list and reaches nothing outside it.
Privacy classes
Privacy evaluation is always applied and cannot be switched off. What counts as private is defined by your organisation, not by ARC.
Spend controls
Rate and cost limits sit on a list and on a person, with organisation-wide limits above both. They are checked while the request is being routed, not reported afterwards.
Fallback paths
When a provider fails, the request takes the next path you set.
Usage analytics
Your token consumption, task-type breakdowns and model usage over time, per key.
See categories and privacy handling

Categories and test bench

A test bench sits beside the rules. Run sample requests through a category and see how it classifies before it touches live traffic.

  • Route normally: No special handling
  • Private destination: Keep it inside the boundary
  • Redact: Remove what the next step does not need
  • Shield: State the task without the material
  • Decompose and recombine: Split the work, reassemble inside
  • Block: Refuse the request
Open the portal
For agents, ARC controls the models, tools, approved data access, policies and spend available to a request. It does not control the agent's reasoning, planning loop, application logic, runtime or business accountability.
Explore model governanceSee where ARC sits in real environments

Control spend and keep billing clear

ARC can prefer less expensive resources among those already allowed and enforce budgets at request time. Access, privacy and required capability remain ahead of cost.

How ARC controls spend

Set budgets that ARC enforces while each request is routed. As a budget fills, ARC can prefer less expensive resources that already meet access, privacy and capability requirements.

How billing is separated

ARC is billed per routing call, with a free tier and no card. Inference remains a direct relationship with your providers, at their rates. FenxLabs takes no margin.

The ARC fee does not change with the destination a request takes.No margin on model inference.
What it costsSee where FenxLabs takes its fee

Where request data sits on the Hosted Platform

Both deployments run the same privacy workflow, and they do not draw the same residency boundary.

Hosted Platform

ARC does not retain or log request data beyond what you configure. On the Hosted Platform the router holds the operational metadata it needs to make a routing decision and to bill the call.

Self-hosted Licence

On the Self-hosted Licence, ARC request handling runs inside infrastructure you control.

ARC controls what each provider is sent. What a provider then retains or trains on is set by that endpoint and its terms, not by the router.

This supports your compliance position. It does not deliver compliance on its own.
See how ARC makes routing decisionsRead the privacy policy

Prove it on one workload

A bounded workload with its own key lets you confirm access, route selection, consumption and configured fallback before production traffic moves.

  1. 01Scope one workloadTake a single bounded use case and give it a key of its own, so what you learn is attributable to it.
  2. 02Define its governed boundaryGive the workload access only to the approved list it needs, with the models, tools and rules that should apply.
  3. 03Review the routesWatch which approved routes policy selects and what they consume. Check the fallback paths you configured, then decide where ARC should run in production.

If access policy, data boundaries or deployment need to be settled before production, bring the workload to an ARC engineer while it is still bounded.

Talk to an engineerChoose where ARC runsRun your own numbers

Start with one governed workload

Start free with one OpenAI-compatible integration. Talk to an engineer if access policy, data boundaries or deployment need to be settled first.

Start freeTalk to an engineer