One API. Every model.
Your spend in plain sight.

Frontier and open-weight models behind one OpenAI-compatible API. Hard budgets that stop spend at the cap, per-request attribution you can trace to the line, and no prompt storage, so cost and data stay under your control. Built on AWS.

Free to join the waitlist. We'll email you as soon as Strata opens to the public.

Already building on Strata

StrataChat StrataBench StrataSwarm Garden ContextEngine VibeCheck

Everything you need, in one gateway.

OpenAI-compatible API

Strata speaks the OpenAI Chat Completions and Responses APIs, including streaming, tool calls, and multi-modal content where the underlying model supports it. Swap one URL and one key. Keep every line of SDK code you already wrote.

Multi-model routing

One key reaches a curated catalog: leading frontier models (Anthropic Claude, OpenAI GPT) alongside open-weight models (Google Gemma, Mistral, and more). Mix and match inside a single app without juggling SDKs or accounts.

Keys, tags, and attribution

Mint as many API keys as you need, one per service or environment. Attach freeform tags (project, cost center, feature) to every request. Slice usage by any tag combination in the dashboard or via API.

Hard budgets, real ceilings

Set monthly caps per key or per user. When a request would exceed the cap, Strata returns 429 immediately instead of letting the bill spiral. Budgets are checked server-side on every call, not in next-day reports.

StrataSwarm deep research

A multi-agent orchestrated deep research platform. An orchestrator spawns specialised workers, each with their own context and tools, and coordinates them to investigate, synthesise, and produce structured reports. Every run captures a virtual filesystem, a todo board, and a replayable execution trace.

Chat + benchmarking, included

StrataChat is a thread-based chat UI with skills and project memory, useful for non-engineers on your team. StrataBench runs identical prompts across multiple models side by side so model selection is data-driven, not vibes-driven.

Works with the tools you already use.

Anything that speaks the OpenAI-compatible API points at Strata with one base URL and one key. The tools below are just a few examples, editors, chat interfaces, agents, or your own application all connect the same way.

Cursor

Point your AI editor at Strata and code against any model on one key.

VS Code

Wire AI extensions to Strata for governed, budget-capped completions.

Android Studio

Give the in-IDE assistant a single, attributable gateway.

Open WebUI

Stand up a private chat front end, or any off-the-shelf chat UI, in minutes.

…and many more. If it can target the OpenAI API, it works with Strata.

Or build your own

Customer-support copilot

Answer tickets and draft replies, with per-team budgets so support spend never surprises finance.

Internal knowledge assistant

Retrieval-augmented Q&A over your own docs and wikis, every request tagged and traceable.

Multi-model router

Send each request to the cheapest model that passes your evals, and switch models without touching app code.

Point your SDK and go.

# Python. OpenAI SDK pointed at Strata.
from openai import OpenAI

client = OpenAI(
    base_url="https://stratablock.io/v1",
    api_key="YOUR_STRATA_KEY",
)
resp = client.chat.completions.create(
    model="claude-opus-4.8",
    messages=[{"role": "user", "content": "Hello!"}],
    # tag the request for usage attribution
    extra_headers={"X-Strata-Tags": "project=onboarding,env=prod"},
)
print(resp.choices[0].message.content)

// TypeScript. openai-node pointed at Strata.
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://stratablock.io/v1",
  apiKey: "YOUR_STRATA_KEY",
});
const resp = await client.chat.completions.create({
  model: "claude-opus-4.8",
  messages: [{ role: "user", content: "Hello!" }],
}, { headers: { "X-Strata-Tags": "project=onboarding,env=prod" } });
console.log(resp.choices[0].message.content);

# cURL. Plain HTTP, no SDK required.
curl https://stratablock.io/v1/chat/completions \
  -H "Authorization: Bearer YOUR_STRATA_KEY" \
  -H "X-Strata-Tags: project=onboarding,env=prod" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-opus-4.8", "messages": [{"role": "user", "content": "Hello!"}]}'

// Go. openai-go pointed at Strata.
client := openai.NewClient(
    option.WithBaseURL("https://stratablock.io/v1"),
    option.WithAPIKey("YOUR_STRATA_KEY"),
    option.WithHeader("X-Strata-Tags", "project=onboarding,env=prod"),
)
resp, _ := client.Chat.Completions.New(ctx, openai.ChatCompletionNewParams{
    Model:    "claude-opus-4.8",
    Messages: []openai.ChatCompletionMessageParamUnion{openai.UserMessage("Hello!")},
})
fmt.Println(resp.Choices[0].Message.Content)

FAQ

What does "OpenAI-compatible" actually mean here?

Strata implements the OpenAI Chat Completions and Responses APIs, including streaming, tool / function calls, structured outputs (JSON schema), and multi-modal content where the upstream model supports it. Any client library that targets api.openai.com works against your Strata base URL with no code change beyond the URL and the API key.

Which models are supported?

A single endpoint reaches a curated catalog of frontier models (Anthropic Claude Opus, Sonnet, Haiku; OpenAI GPT) alongside leading open-weight and domain-specific models (Google Gemma, Mistral, and more). The catalog grows over time, and the available list is curated per deployment, so your admin decides which models developers can call.

How do API keys and tags work together?

You can mint any number of scoped API keys, typically one per app, service, or environment. On every request you can attach a freeform tag string (via the X-Strata-Tags header) such as project=growth,env=staging. The usage dashboard then lets you group and filter spend by any combination of key, model, and tag, so attribution down to a specific feature flag or experiment is one click away.

How are rate limits and budgets enforced?

Per-key request and token rate limits sit in front of every upstream provider, so a runaway loop in one app cannot starve the others or exhaust your shared upstream quota. Monthly budgets are checked server-side on every call: once a key or user hits its cap, new requests return HTTP 429 immediately. You see the breach in the same request, not in tomorrow's billing email.

Where does my data live? What gets stored?

Strata runs on AWS, so your traffic stays inside AWS data centers from the moment it hits our gateway until it reaches the model and the response comes back to you. We do not store the contents of your requests or responses at all. The only data we persist is operational metadata for monitoring and attribution: model name, input and output token counts, latency, status code, and the tag string you attached. No prompts. No completions. Nothing of your actual workload.

Is there a plan for my team or company?

Yes. Bringing Strata to a whole team or company has its own page: Strata for Organization. You get shared key management, per-person budgets with hard spend limits, role-based access, spend tracked by model, person, and day, and one consolidated bill, without anyone on your side hosting or operating anything. Head there to tell us about your team.

Do you offer a managed service or self-hosting for organizations?

Both. On the Strata for Organization page you can run Strata as a managed service, live in a day with nothing to operate, or deploy it inside your own AWS account for full data isolation. Either way you get hard budgets, spend tracked by model, person, and day, and an audit trail your security team can sign off on. Book a 15-minute call from there.

When can I get access?

We are onboarding in small batches as capacity opens. Joining the waitlist puts you in the queue, and we will reach out by email when your account is ready.

One API. Every model. Your spend in plain sight.