Agent-first · One universal API

One API
for everything you run.

Over 450 models, every tool and connector, managed agents, workflows and bots.

Twelve planes behind one key, with the price on every response.

  • Prices public
  • Price on every response
  • Spend caps
  • Approved providers only
  • Prompts not kept
  • Public status
  • Free to use
  • Pay per result
  • 12 planes · 1 server

One call per plane

Everything behind the key

From one model call to a fleet of agents, on one account, one request shape and one bill.

Read the guide
  • POST /v1/chat/completions — 200 · 0.8 s
  • POST /v1/tools/serper-agent/execute — 200 · 1.9 s
  • POST /v1/agents/invoice-chaser/execute — spent $16.02 of $25
  • POST /v1/workflows/execute — 200 · 4 tasks
  • POST /v1/workflows?kind=bot — 201 · live in #support
  • GET /v1/discovery/search — 3 agents · 12 tools

Every plane is free to use. You pay per result.

Models

One request shape for every model

Name a model and list fallbacks. When a provider fails, the call moves to the next one, and the response says who served it.

How routing works

Pricing

The price is on every response

Each call returns its cost. Give an agent a monthly budget, and it stops at the cap instead of at the invoice.

Pricing

Agents

Agents you can see and stop

One list of every agent, workflow and bot: who owns it and what it spends. Pause one, or stop them all with one switch.

Managed agents

The routing planes

Twelve planes. One endpoint.

/models

Model routing

Over 450 models, routed on cost, latency and quality, with failover

/mcp

MCP routing

Over 1,000 MCP servers behind one endpoint

/agents

Managed agents

Over 1,300 agents, each a bounded mission with an owner and least privilege

/bots

Bot agents

Slack, WhatsApp and web bots on a schedule

/tools

Tool routing

Search, code, browsers: one schema, every tool

/skills

Skill routing

Reusable capabilities, injected per run

/workflows

Managed workflows

Deterministic multi-agent runs

/payments

Payments (x402)

Machine-to-machine, settled per result

/connectors

Connector routing

Over 1,000 connectors, one auth

/plugins

Plugin routing

Tool packs, resolved at run time

/long-running

Long-running agents

Hours or days, with checkpoints and memory

/discovery

Discovery

Agents, models, tools and connectors, searchable

Read the docs

Over 450 models, over 1,000 connectors and over 1,300 agents behind one key.

Over 450

Models

Routed on cost, latency and quality, with failover.

Over 1,000

Connectors

Over 1,000 MCP servers behind one endpoint.

Over 1,300

Agents

Each a bounded mission with an owner and least privilege.

12

Planes

Twelve planes behind one key.

Works with what you use

  • OpenAI SDK
  • Anthropic SDK
  • OpenRouter clients
  • MCP clients
  • curl

For startups

Ship in an afternoon

One key for over 450 models, every tool and connector. Agents from a playbook, not a platform team.

Guide

For small businesses

No platform team needed

A bot in Slack or on your website in a day. Each agent stops at its budget, never at the invoice.

See the bots

For enterprise developers

Hundreds of APIs become one

Replace integration sprawl with one gateway. Budgets, permissions and audit on every call.

API reference

For platform teams

Governed from birth

Identity, owner and policy for every agent at creation. Pause one, or stop them all with one switch.

Agent control

Trust

Built for production, with nothing hidden

What a security review asks first, answered where you can check it.

  1. Prices in the open

    Every model’s price, readable without a key. A posted price is never above the maker’s own list price.

    GET /v1/pricing
  2. The price of every call

    In a response header, in dollars, or looked up afterwards by the call’s id.

    x-ocean-cost-usd
  3. Spend caps

    For the organisation by day, week or month, and per run or per month on each agent. A call past the cap is refused.

    429 spend_cap_exceeded
  4. Only the providers you approve

    Allow-list the models and providers you trust, or require your own provider keys on every call.

    require_byok
  5. Prompts are not kept

    The model request log holds counts, cost and timing, never the prompt or the answer.

    Docs › Data and privacy
  6. Status in public

    Each provider’s last 24 hours, open to anyone, and a request id on every call.

    gateway.plungeai.com/status

Not yet

A SOC 2 report or an ISO 27001 certificate. Data and privacy says what is in place today; Status shows each provider live.

Agent

Read this page as an agent

# PlungeAI One API
> One API for everything you run: models, tools, agents, workflows, bots.

- Base URL: https://api.plungeai.com/v1 · Auth: Authorization: Bearer <key>
- Models: POST /v1/chat/completions · GET /v1/models (over 450)
- Tools: POST /v1/tools/{id}/execute · Agents: POST /v1/agents/{id}/execute
- Workflows: POST /v1/workflows/execute · Bots: POST /v1/workflows (kind: bot)
- Search: GET /v1/discovery/search · Prices: GET /v1/pricing (no key needed)
- MCP: https://mcp.plungeai.com/v1 · Cost of each call: x-ocean-cost-usd header

Make your first call in two minutes

Base URL, one key, and the SDK you already use. Or connect your tool over MCP.

Want it done for you instead? Our own Agency and Consulting teams build on this API, and Ocean Code writes the code.

Agency Consulting Ocean Code