v2.14 · canary promotions shipped

The control plane for production AI.

Strataix sits in front of Anthropic, OpenAI, Google and others — records every request, surfaces the quiet waste, and cuts it. Routing, guardrails, budgets and traces in one gateway.

LIVE FINDING
Cache hit 84% → 12% on support/triage after a prompt reorder · restoring saves ~$860/day.
Gateway · 24h flowlive
YOUR APP
support/triage
copilot/complete
data/extract
chat/assistant
STRATAIX · GATEWAY
route
guardrail
cache
fallback
budget
trace
replay · A/B · audit
PROVIDERS · 5
Anthropic
OpenAI
Google AI
Fireworks
Mistral
1.24M requests · 24h $4,182 spend p95 1.84s 61.4% cache hit
1.24Mrequests routed · 24h
5providers behind one API
61%of calls served from cache
99.95%gateway uptime · 30d
01 · WHAT THE LOG REVEALS

Every request tells you something. Most teams throw that log away.

Three findings surfaced from one week on acme · all solvable by changing config, not code.

CACHE BREAK
$860/day recoverable
A prompt reorder on support/triage dropped hit rate 84% → 12%. Order restored in one commit.
OVER-TOOLED
29.9ktokens per call pruned
31 unused tools were loaded every turn on copilot/complete. Pruning them saves ~28% of every input.
CANARY WIN
17%p95 faster · 7% cheaper
A sonnet-4-5 canary beat the baseline on both metrics. Promoted with one click.
02 · THE CONTROL PLANE

One gateway. Six layers of control.

Everything you'd otherwise rebuild in-house — routing, guardrails, budgets, traces, keys, audit — already here.

Shipping

Routes · primary + fallbacks + A/B

Point your SDK at Strataix. Define a route with a primary model and ordered fallbacks on 5xx / 429 / timeout. Promote a canary with one click.

5xx fallbacktimeout retryA/B splitcanary promote
Shipping

Guardrails · policy before it leaves your network

PII redaction, prompt-injection detection, toxicity, secret-leak detection — each as a toggle per route. Applied with +14ms p95.

PII redactprompt injectiontoxicitysecrets
Shipping

Budgets

Caps by team, route or key. Warn → page → downgrade → hard stop on a ladder you set.

Shipping

Traces

Every tool call and LLM call as a span. Replay with new inputs. OTel export.

Researching

MCP studio

Browse installed MCP connectors, test tool calls, expose them per route.

03 · HOW IT WORKS

Point your SDK at a different URL. Everything else stays yours.

Keep your Anthropic / OpenAI SDKs. Keep your enterprise discounts. Keep your code. Strataix adds the dashboard, the routing, and the guardrails around them.

01

Connect providers

Paste your Anthropic / OpenAI / Google key once. Stored encrypted, scoped per environment. We proxy at their list price.

02

Issue a Strataix key

Narrow scopes, per-route rate limits, 90-day rotation. Point your existing SDK at gateway.strataix.in/v1.

03

Ship the dashboard

Your first request lands in the log in <1s. Open a route, add a budget, turn on guardrails — then forget you installed it.

04 · WHAT'S SHIPPED, WHAT'S NEXT

We're honest about what's here and what isn't.

Strataix is a small, focused product. Below is the state of each core capability today.

NOVEMBER 2026

Routing, guardrails, budgets and traces are in production on 5 providers.

Key leak detection ships this month. MCP studio is a research track — we're prototyping with 3 design partners before we commit to an API.

Not here yet: mobile dashboard, SIEM export, HIPAA attestations. If any of those block you, write to us and we'll scope it with you.

SHIPPED13
routing · guardrails · budgets · traces · audit · SSO · Slack · webhooks · keys · providers · billing · alerts · settings
RESEARCHING3
MCP studio · Agent SDK traces · on-prem

Ship with the providers you choose. Keep the one gateway in front.

Free for the first 50k requests a month, forever. No card. First request flowing in under 4 minutes.