AI API gateway
OpenAI-compatible
Defense planned

Run any model. Stop rogue agents.

Text and audio models through one OpenAI-compatible API at launch, with image, video and Myriax Defense planned.

Join early access See what ships at launch

Pre-launch. Early access opens to US teams first.

One request, traced

See where every millisecond and cent goes.

Each call through Myriax will be traceable span by span: the Defense check, the budget, the route, the model and the receipt. This one is an illustrative sample.

TRACE req_8f31c2

Duration: 480 ms
Spans: 6
Defense: allow
Upstream: provider
Quoted: $0.000412

upstream

The model call, routed to the upstream the route picked.

upstream	provider
waiting	130ms
streaming	316ms
charged_usd	0.000398

Illustrative trace with sample values. The API is not live yet.

How it works

One gateway between your code and every model.

At launch, requests route to the model you pick. Defense is planned to check traffic at your door and the agents inside your stack.

Illustrative · sample data

Illustration: your code sends each request to the Myriax gateway, which checks it against Defense rules, routes it to a model provider and returns the answer. Agent tool calls pass through the same gateway.

Request Illustrative

curl https://api.myriax.ai/v1/chat/completions \
  -H "Authorization: Bearer $MYRIAX_KEY" \
  -d '{"model": "deepseek/deepseek-v3.2",
       "messages": [{"role": "user",
                     "content": "Hello"}]}'

Illustrative. The API is not live yet.

Route

endpoint: POST /v1/chat/completions
model: deepseek/deepseek-v3.2
upstream: provider
defense: allow

Receipt Sample

receipt: rcpt_01JBX4K9
event: settled

Signed receipts are under validation.

Request rules Sample

Path traversalblocking
Scanner bursts on /.env, /.gitblocking
Spoofed crawlersblocking
SSRF to cloud metadataobserve-only
Signed agentsallow

Every new rule will start in observe-only mode, so you see what it would block before it blocks anything.

Rules recognise attacks. They don't patch your app, so keep patching.

Kill switch

Through the gateway it will apply to the next request.

Hard spend limit Sample

support-bot/prod · twenty replicas, one budget

Budgets will be reserved before each model or tool call.

Audit log Sample

Planned: enforced at the Myriax gateway for calls routed through it, and in the SDK for wrapped tools.

Models at launch

Every task, one compatible API.

One OpenAI-compatible API, organised by task, so you can pick by the job you need done. Launch models depend on signed provider agreements.

01 / TEXT

Text generation

At launch

POST /v1/chat/completions

02 / STT

Speech to text

At launch

POST /v1/audio/transcriptions

03 / TTS

Text to speech

At launch

POST /v1/audio/speech

04 / IMAGES

Image generation and editing

At launch once image safety screening is approved

POST /v1/images/generations, /edits

05 / VIDEO

Video

Not at launch

POST /v1/jobs

06 / CATALOG

See the planned models

Myriax Defense · planned

Two sides of the same problem.

Agents now hit your API from both directions. Defense will block the ones attacking you and limit the ones you run.

At your door

Block hostile agents

Built to stop exploit bots and spoofed crawlers before they reach your site or API.

Request log Sample

block GET /.env scanner probe
allow GET /docs signed agent
limit POST /v1/chat spend limit
block GET /fetch?u=169.254.169.254 SSRF to metadata
observe GET /../../etc/passwd new rule

block: 2
limit: 1
allow: 1
observe: 1

Illustrative. Rules recognise attacks; they don't patch your app, so keep patching.

Inside your stack

Keep your agents in bounds

Your own agents will get permissions, a budget and a switch that stops them.

agent: support-bot Running

Kill switch

tools

approval issue_refund needs a human

budget $3.80 of $10.00

egress api.helpdesk.example

Illustrative. Planned: enforced at the Myriax gateway for routed calls, and in the SDK for wrapped tools.

How Defense works

Under validation

Two ideas we'll build if teams want them.

We build each one only if enough early-access teams say they'd use it.

M1 / RECEIPTS

Signed receipt chain

A signed record of each call's route, declared model, price and Defense decision, checkable offline.
M2 / RE-APPROVAL

Re-approval on change

If a pinned route's model version, upstream or price changes, calls would wait until you approve the diff.
See the samples

Pricing · planned, early access

Pay for what you run.

01 / MODELS

Pay as you go

$10 minimum top-up

Billing: Prepaid credits. No subscription.
Text models: Provider cost plus 2%, rounded up per request.
Platform fee: 4.9% on each top-up ($0.80 minimum)

02 / DEFENSE

Myriax Defense

$0 a month on the free tier

Included: 25,000 requests and 10,000 agent actions a month.
Paid plans: From $9 a month. We're setting final plan prices with early-access teams.

Planned prices for early access, not an offer. US cards and billing addresses at launch. Full pricing

Early access

Join early access

We're onboarding a small group of teams first. Tell us what you'd use.

Preview: this form isn't connected.

Early access is currently for United States residents age 18 or older acting for a business or team.

Work email

Team or company optional

What would you use?

Models Defense Both

Would you use either idea we're testing? optional

Signed receipt chain Re-approval on change

If Myriax Defense came in one of these plans, which would you pick? optional

We're testing two price ladders with the same features. Reference prices, not an offer.

I confirm that I meet the eligibility requirements and agree to the Terms of use.