Which intelligence?
Capability, task fit, context, modality, tools, quality, and risk.
MC-1 · Adaptive Intelligence Control Plane
Models generate intelligence.
MC-1 controls it.
MC-1 governs how applications, agents, and autonomous systems access, compose, evaluate, secure, and continuously improve intelligence across local, open, specialized, and frontier models.
Continue with email, Google, or an existing ChatGPT account.
Production control plane · Open the console to execute
INTELLIGENCE LIFECYCLEGive MC-1 the outcome.
Give MC-1 the outcome.
It governs the intelligence.
The production console turns task, risk, identity, policy, budget, model, provider, evaluation, and escalation controls into one auditable execution.
01 REQUEST
02 INTELLIGENCE TRACE
1
Understand taskIntent, complexity, domain, and risk
2
Verify identity and policyAuthorization before optimization
3
Allocate intelligence budgetBalanced objective within customer limits
4
Compose model and provider routeEligible local, managed, BYOK, or specialized execution
5
Evaluate and correctBounded verification, retry, and escalation
03 ROUTE DECISION
LIVE ONLYEvery decision
is request-specific.
No fabricated route or benchmark data.
- Objective
- Balanced
- Policy
- Evaluated at execution
- Escalation
- Customer-controlled
- Evidence
- Recorded in Intelligence Trace
The best intelligence
is contextual.
No single model leads across every task, modality, budget, region, and privacy boundary.
MC-1 operates one layer above individual models—continuously turning constraints into an executable route.
STATIC SELECTION
1
One model
for every task
→
MC-1 ORCHESTRATION
RCVAMMC-1
Dynamic intelligence
for every request
CONTROL PLANERoute when routing is best.
Route when routing is best.
Adapt when specialization is better.
MC-1 compares measured quality, latency, cost, volume, authorization, and lifecycle risk before recommending routing, an experiment, or user-owned specialized intelligence.
01Understand→
02Select→
03Execute→
04Guard→
05Evaluate→
06Learn→
07Optimize
One policy-aware decision.
Constraint filtering, utility optimization, confidence, and escalation.
Where should it run?
Cost, latency, health, privacy, region, quota, and availability.
MODEL INTELLIGENCEKnow what every model
Know what every model
is actually good at.
A canonical registry turns provider-specific model IDs into comparable capabilities, constraints, and evidence. Numerical scores only appear with provenance.
CANONICAL PROFILE · SCHEMA
Reasoning specialist
Provider-agnostic model identity
Task fitPopulated only from provenance-backed evidence
ReliabilityPopulated only from provenance-backed evidence
Tool supportPopulated only from provenance-backed evidence
Context fitPopulated only from provenance-backed evidence
- Streaming
- Supported field
- Regions
- Provider-specific
- Pricing
- Dynamic input
- Provenance
- Mandatory
SECURE AGENTIC INTELLIGENCEIdentity and policy
Identity and policy
before execution.
Agent Identity, Agent Guard, and policy-aware execution constrain sensitive tools and high-risk actions before provider inference begins.
Know who is requesting intelligence.
Tenant, principal, trust attributes, certification, permissions, and expiration travel with agentic requests.
Authorize the execution.
Privileged tool use is checked against identity and permissions before cost is incurred.
Did the task actually succeed?
Deterministic format, tool-call, and completion checks run before bounded recovery.
EXECUTION PLANEYour models. Your keys.
Your models. Your keys.
Your compute.
MC-1 is designed as the intelligence and control layer. Inference can remain with managed providers, a private cloud, or infrastructure you control.
EXECUTION TARGETIMPLEMENTATION STATUSLOCATION
Customer / BYOK computePreferred when eligibleCustomer-defined
Nebius Token FactoryDirect adapter readyProvider-defined
Together · Fireworks · NIMDynamic managed routingProvider-defined
OpenRouter networkBroad coverage / resilienceProvider-routed
Statuses describe this interface build and adapter architecture—not live commercial availability. Production credentials and provider validation are required before traffic is routed.
Keep your contracts.
Add routing intelligence.
Connect existing provider credentials, define eligibility, and let MC-1 select among approved routes.
Your provider↓Your credentials↓MC-1 routing↓Your compute
DEVELOPERSReplace model selection
Replace model selection
with one endpoint.
The interface follows OpenAI-compatible chat completion patterns. Switch the base URL and use colomboai/mc-1.
from openai import OpenAI
client = OpenAI(
base_url="https://api.colomboai.com/v1",
api_key="$COLOMBOAI_API_KEY"
)
response = client.chat.completions.create(
model="colomboai/mc-1",
messages=[{"role": "user", "content": "Build this..."}]
)MODEL ALIAScolomboai/mc-1
PRIMARY ENDPOINT/v1/chat/completions
INTERFACEOpenAI-compatible
ORGANIZATIONAL CONTROLIntelligence without
Intelligence without
infrastructure lock-in.
Organizations define which models may run, where data may flow, what a request may cost, and how execution is audited.
MC-1 for
Enterprise
- BYOK and private endpoints
- Model and provider allowlists
- Budgets, observability, audit logs
- Custom routing and deployment
MC-1 for
Government
- Sovereign and region restrictions
- Policy-controlled routing
- Customer-managed infrastructure
- Auditable execution architecture
1Application
2MC-1
3Policy
4Approved models
5Approved infrastructure
6Audit
Designed to support security-controlled and sovereign AI deployment architectures. No certification or authorization status is implied.
TRANSPARENT PRICINGOne network. Every intelligence layer.
One network. Every intelligence layer.
Transparent pricing.
Model inference stays at the published provider or network price. MC-1 adds a 4.9% Network Fee when credits are purchased; Smart Routing, manual model choice, and Provider Intelligence are included.
MODEL INFERENCEProvider / network prices
MC-1 NETWORK4.9%
SMART ROUTINGIncluded
MANUAL MODEL CHOICEIncluded
$0/month
- MC-1 API and playground
- Manual model access
- Limited MC-1 Smart
- Basic Model Intelligence
- One project
- Community support
$20/month
- Everything in Free
- Higher API limits
- Up to five projects
- 30-day activity history
- Advanced analytics
- Developer adaptation tools
$99/month
- Everything in Developer
- Higher API limits
- Advanced routing controls
- Advanced usage analytics
- Specialized-model workflows
- 90-day activity history
$499/month
- Everything in Pro
- Team workspaces
- Organization policies and RBAC
- Shared BYOK
- Shared specialized models
- Agent Identity and Agent Guard management
- Audit logs
- Training and adaptation workflows
- One-year activity history
$2,500/month
- Everything in Team
- Advanced governance
- ChatGPT identity access
- Higher project and API limits
- Advanced Agent Guard
- Adaptation and training intelligence
- Seven-year activity retention
First $25,000/month included,
then 4.0%
Enterprise BYOK includes the first $250,000/month equivalent inference, then 3% or contracted volume pricing.
Only when you use it.
Evaluation starts at $0.003. Additional model branches are $0.01. Agent Guard starts at $0.005 per guarded action. Activation requires a billing contract until atomic usage collection is enabled.
From $50,000/year.
MC-1 Government starts at $250,000/year. Scope, deployment, support, and sovereignty requirements are priced contractually and transparently.
BENCHMARKSIntelligence should be
Intelligence should be
measured as a system.
MC-1 separates measured results, third-party evidence, and projections. Empty evidence is shown honestly rather than replaced with marketing numbers.
QUALITY ↑
COST →
No measured pointsThe frontier appears only after a controlled benchmark is recorded.
Evidence states
MeasuredMC-1 controlled benchmark
Third-partyCited external result
ProjectedExplicitly modeled estimate
No production benchmark dataset is connected to this build.
COLOMBOAI RESEARCH · PREPRINTAdaptive, efficient,
Adaptive, efficient,
and sovereign.
ColomboAI-MC-1: A Mixture-of-Models Intelligence System for Adaptive, Cost-Efficient, and Sovereign AI Inference
The paper moves conditional computation from the intra-model level of Mixture-of-Experts to the inter-model level: MC-1 routes requests—and, when useful, reasoning stages—across independently trained models using capability, cost, latency, privacy, policy, availability, and confidence.
10 pages6 principal contributions2025 development began
PREPRINT · AUGUST 2026A Mixture-of-Models
MC
—1
—1
A Mixture-of-Models
Intelligence System
Architecture, mathematical formulation, systems design, evaluation methodology, and a research agenda for programmable collective intelligence.
- Evaluation
- Quality · cost · latency
- Router metric
- Oracle recovery
- Claim boundary
- Architecture & methods
Kouadio, W., & Li, A. (2026). ColomboAI-MC-1: A Mixture-of-Models Intelligence System for Adaptive, Cost-Efficient, and Sovereign AI Inference. Preprint. ColomboAI, Cairo Lab.
Build the intelligence
layer with us.
Infrastructure providers and model labs can become eligible participants in the MC-1 ecosystem after integration and evaluation.
Become an MC-1
inference provider.
Make your infrastructure eligible to receive intelligently routed workloads.
Integrate your infrastructure →Add your model
to MC-1.
Submit a model to the Model Intelligence evaluation pipeline.
Submit a model →Questions, answered.
MC-1 extracts request features, applies tenant policy, generates eligible model and provider candidates, ranks routes against the chosen objective, executes, evaluates, and escalates within explicit limits.
Stop choosing
models.
Give MC-1 the task. It will find the intelligence.
The future of open AI isn't one model. It's a Mixture of Models.