ONE/API
Universal AI Control Plane · v4.0
One gateway · One marketplace · One control plane

The One Platform
for all of AI.

Four ways in — Studio, Agent Desk, API, MCP. One gateway in the middle. Every model, tool, connector, MCP server, skill, plugin, and agent on the other side — discoverable, routable, manageable.

model routing tool routing connector routing mcp routing skill routing plugin routing agentic agents agent flow discovery service
Fig. 01 — The routing fabric
v4.0
ENTRY POINTS GATEWAY ROUTING PLANES studio build visually agent desk operate agents api code-first mcp protocol-native ONE API route · auth · meter ⇄ models every LLM · per token tools functions · search · code connectors 500+ SaaS integrations mcp servers official · community · private skills reusable know-how packs plugins bundled capability installs agentic agents autonomous · agent ↔ agent agent flow orchestrated workflows discovery the registry of everything
01 / Entry points

Four ways in. Same gateway.

Every audience meets the platform on its own terms — but all traffic flows through the same One API, so auth, billing, observability, and governance are identical no matter where a request starts.

in-01 · visual

Studio

The visual builder. Compose agents and workflows by wiring models, tools, and connectors together — no code required.

in-02 · operate

Agent Desk

The operations workspace. Monitor, steer, and approve running agents — a human-in-the-loop console for everything in production.

in-03 · code

API

The developer path. One REST/SDK endpoint that speaks every plane — drop-in compatible, from prototype to production.

in-04 · protocol

MCP

The protocol-native path. The entire platform exposed as a single MCP server — any MCP client gets all planes instantly.

02 / The problem

AI infrastructure is six separate stacks.

Every AI product today stitches together fragmented layers — each with its own vendors, keys, billing, and dashboards. Nobody manages them in one place.

Models
Separate keys, quotas, prices, and failure modes per provider.
Tools
No standard catalog. No unified access or quality signal.
Connectors
Locked in opinionated iPaaS platforms that own your workflow.
MCP servers
Hundreds exist — but no routing, auth, or trust layer in front.
Agents & flows
Agent-to-agent traffic has no gateway, no orchestration standard, no marketplace.
Skills & plugins
20,000+ skills and thousands of plugins scattered across GitHub marketplaces — no unified registry, versioning, or trust.
Agents are the workload that breaks this model: a single agent request is a model call + tool calls + connector calls + calls to other agents — today managed across four or five different systems.
03 / The solution

Nine planes. One endpoint.

The same pattern OpenRouter proved for models — applied to the entire AI supply chain. We don't build any of it. We route, meter, and manage all of it.

route /models

Model Routing

Every LLM — frontier, open-source, fine-tuned — routed on cost, latency, and quality. Automatic failover and load balancing across providers.

analogy: OpenRouter
route /tools

Tool Routing

Every tool and function endpoint — search, code execution, browsers — routed on capability and rating. One schema, every tool.

analogy: no incumbent — open field
route /connectors

Connector Routing

500+ SaaS integrations from any provider — Composio, Merge, Nango, or native APIs. Neutral aggregation: we don't own the workflow, we expose the integration.

analogy: unified API, but neutral
route /mcp

MCP Routing

Every MCP server — official, community, enterprise-internal — behind one endpoint that handles auth, selection, and versioning. The protocol where tools and connectors converge.

analogy: MCP gateway
route /skills

Skill Routing

Reusable expertise as endpoints: instruction packs and playbooks — versioned, rated, and loaded by any agent on demand. Tens of thousands exist today, scattered across GitHub directories.

analogy: npm for know-how
route /plugins

Plugin Routing

Installable capability bundles — skills, commands, connectors, and MCP servers packaged as one versioned unit — one-click install into any agent or workflow.

analogy: app store for agent capabilities
route /agents

Agentic Agents

Autonomous agents addressable as endpoints. Agent-to-agent (A2A) traffic with a trust layer, metering, and a marketplace — a plane nobody has claimed yet.

analogy: app store × API gateway
route /flows

Agent Flow

Orchestrated multi-agent workflows: sequences, hand-offs, and human checkpoints — composed in Studio, operated from Agent Desk, invoked like any other endpoint.

analogy: workflows as endpoints
service /discovery

Discovery Service

The registry across all planes: search, ranking, trust scores, and versioning for every model, tool, connector, MCP server, skill, plugin, agent, and flow. Agents don't just call what they're configured with — they find what they need at runtime.

analogy: DNS + npm for AI
04 / The control plane

Routing is the wedge. Management is the moat.

One key, one bill, one dashboard — the durable value of being the single pane of glass for all AI traffic, no matter which entry point it came through.

C-01

Unified auth

One credential replaces fifty. OAuth, keys, and scopes managed centrally.

C-02

Unified billing

One invoice across model tokens, tool calls, connector usage, and agent invocations.

C-03

Observability

End-to-end traces spanning the model call and every tool, connector, and agent call it triggered.

C-04

Governance

Policy on which teams use which models, tools, and data. Audit trails and compliance built in.

C-05

Reliability

Failover, load balancing, and caching at every plane — not just the model layer.

05 / The landscape

Nobody spans all nine planes.

Player Models Tools Connectors MCP Skills Plugins Agents Flows Discovery
OpenRouter
LiteLLM / Portkey
Zapier / Merge
Composio
MCP registries
Plugin marketplaces
ONE/API
full coverage partial none
06 / The marketplace

Two sides, one flywheel.

Supply

Model providers, tool builders, connector providers, MCP authors, skill & plugin creators, and agent & flow developers list into the discovery registry — monetized per token, per call, or per seat. We are their distribution channel.

Demand

Builders arrive through Studio, operators through Agent Desk, developers through API, and protocol clients through MCP — all consuming the same catalog through one gateway.

more supply → better routing decisions → more demand → more supply
discovery ranking (quality · latency · price · trust) becomes the moat
The positioning

The One API for AI.

Every model. Every tool. Every connector. Every MCP server. Every skill. Every plugin. Every agent. Every flow.
Studio · Agent Desk · API · MCP — in front
One key · One bill · One gateway · One marketplace · One control plane — behind

We don't build models, connectors, or agents. We are the routing fabric that makes all of them discoverable, interchangeable, and manageable — through one API.