For Builders · AI Gateway · Proof of Decision

OmniStrat Foundry.

The best AI gateway is the one you can govern and audit. That’s OmniStrat AI. One endpoint, every model, every call cryptographically signed. Cost caps that stop, not warn. Failover fast enough it never wakes you. Taskforces that hold a charter, a budget, and a memory.

Usage

Your usage · live
AI calls
.
Signed by Ed25519
Tokens routed
.
Across 17 US + allied providers
Est. cost
.
Under your cap
Taskforces
.
Charter · budget · memory

Your builds

Saved on this device

Your taskforces

Console

Manage your Foundry
API Keys

Generate, rotate, revoke.

Long-lived org keys or short-lived JWTs, provisioned from your console. Rotation is a clean cutover and every rotation lands in your audit chain.

Taskforces

Build agents that hold charter.

Each Taskforce has a charter, budget, memory that compounds, and an approval workflow you configure per action class.

Cost governance

Hard caps, per-org · per-key · per-call.

Set a monthly ceiling. The Router refuses the 601st dollar. No overages. No surprises. Anomaly detection with drift alerts.

Audit chain

Every call receipted.

Every request, signed. Every response, hash-chained. Check any receipt yourself, in your browser, without an account.

Integrate

Use it from your code, or your AI client.

Browse every endpoint and try calls in the browser. Or connect OmniStrat to Claude, Cursor or any MCP client by pasting one URL.

Browse the API
Observability

Ship your telemetry anywhere.

Export spans and structured logs to your own OTLP collector, and stream webhook events on routing, spend and approval decisions.

Quickstart

One line to your first call
POST https://api.omnistrat.ai/api/route
# Bash, native endpoint (credentials go in the body)
curl -X POST https://api.omnistrat.ai/api/route \
  -H "Content-Type: application/json" \
  -d '{ "org_id": "org_...", "api_key": "ost_...", "prompt": "Hello" }'

# -> { ok, provider, model, result, tokens_out,
#      receipt: { receipt_id, entry_hash, verify_url } }
# Every response carries a receipt anyone can verify, no account.

# Already using OpenAI? Change one line.
import OpenAI from "openai";
const client = new OpenAI({
  apiKey:  "ost_...",
  baseURL: "https://api.omnistrat.ai/v1",
});
const r = await client.chat.completions.create({
  model:    "auto",
  messages: [{ role: "user", content: "Hello" }],
  org_id:   "org_...",   // OmniStrat: your org
});

Providers behind the endpoint

17 available · US + Western-allied · auto-failover
Anthropic
US · San FranciscoClaude Opus · Sonnet
OpenAI
US · San FranciscoGPT-5 · GPT-4o
Google
US · Mountain ViewGemini 2.5 · 1.5
xAI
US · Palo AltoGrok 4 · 3
AWS Bedrock
US · SeattleClaude · Llama · Titan
Azure OpenAI
US · RedmondGPT-5 via Azure
Cloudflare AI
US · San FranciscoWorkers AI · Llama · Mistral
Groq
US · Mountain ViewLPU inference
NVIDIA NIM
US · Santa ClaraLlama · Nemotron
Replicate
US · San FranciscoOpen-weights host
Perplexity
US · San FranciscoSonar · search-native
Anyscale
US · San FranciscoRay Serve · Llama
Hyperbolic
US · San FranciscoH100 · open-weights
Baseten
US · San FranciscoTruss · fine-tunes
Lepton AI
US · Palo AltoPhoton · open-weights
Mistral
FR · Paris · alliedLa Plateforme · Large
Cohere
CA · Toronto · alliedCommand R · Rerank
Go deeper The LLM router with an audit trail, explained → Code Mode: sandboxed agent execution → How to prove AI decisions →