@omnistrat/sdk

Official TypeScript SDK for the OmniStrat AI Router.

One API. Thirteen providers. Edge-deployed. Cost-capped. BYOK. Audit-chained.

npm install @omnistrat/sdk

Quickstart

import { OmniStrat } from "@omnistrat/sdk";

const client = new OmniStrat({
  apiKey: process.env.OMNISTRAT_API_KEY!,
  orgId:  process.env.OMNISTRAT_ORG_ID!,
});

const r = await client.route({ prompt: "Explain stoicism in three sentences." });
console.log(r.result);

Streaming

for await (const chunk of client.routeStream({ prompt: "Tell me a story." })) {
  process.stdout.write(chunk.text);
}

Routing profiles

Pick how the Router resolves provider + model for you:

await client.route({ prompt: "...", profile: "cost"     });  // cheapest available
await client.route({ prompt: "...", profile: "quality"  });  // best-in-class model
await client.route({ prompt: "...", profile: "latency"  });  // lowest-latency
await client.route({ prompt: "...", profile: "race"     });  // first to respond wins
await client.route({ prompt: "...", profile: "balanced" });  // default

Direct provider + model

await client.route({
  prompt:   "...",
  provider: "claude",
  model:    "claude-opus-4-6",
});

BYOK

await client.attachKey({ provider: "openai", api_key: "sk-..." });
// Subsequent calls bill your provider account, not the platform's.

Agents (long-running)

const agent = await client.agentStart({
  goal:  "Research the Mendel pea experiments and summarize.",
  tools: ["web_search", "summarize"],
});
const result = await client.agentWait(agent.id);

RAG

const answer = await client.rag({
  index: "your-index",
  query: "What is the doctrine of consubstantiation?",
});

Reputation

const me = await client.reputation();
// { score: 87, tier: "trusted", router_score, terminal_score, game_score, billing_score }

Embedded environments

The SDK is zero-dependency and uses only platform fetch + ReadableStream. Works in:

• Node 18+
• Cloudflare Workers
• Deno
• Bun
• Modern browsers

Endpoints covered

• /api/route — chat completions
• /api/route/stream — streaming chat completions
• /api/embed — embeddings
• /api/rag — hybrid retrieval
• /api/moderate — content moderation
• /api/estimate — cost + token estimation
• /api/image — image generation
• /api/transcribe — audio transcription
• /api/speak — text-to-speech
• /api/agent — long-running agents
• /api/memory — agent memory
• /api/reputation/me — cross-product reputation
• /api/byok — provider key attachment
• /api/whoami — caller identity + tier

MCP

The Router exposes every tool through Model Context Protocol at /api/mcp/v1. AI agents discover the surface at /.well-known/mcp/server-card.json.

License

MIT