AI infrastructure for SaaS

Ship AI in your SaaS this afternoon.

Per-user wallets, budgets, rate limits, and tool calls — for any model. Drop in our SDK and your product has AI by the end of the day.

Request access →Read 5-min quickstartinvite processed same-day · no credit card required
// One line. Per-user billing baked in.
import { assistiv } from "@assistiv/sdk";

export async function POST(req) {
  const { messages, userId } = await req.json();

  return assistiv.chat({
    user: userId, // per-user wallet auto-debited
    model: "gpt-4o",    // or claude, gemini, anything
    messages,
  });
}
user_28f3a · liveActive
balance$5.80
budget (mo)$10.00
spent today$0.42
42% usedresets in 18 days

Built for production from the first request

SOC 2 Type IIaudit in progress · DPA on request
MCP-compatibleany MCP server as a tool
99.95% SLAmulti-provider fallback
EU & US residencypick where data lives

Per-user billing, built in

Your users never overspend. You never eat the cost.

Wallets, budgets, and pre-flight cost checks per end user — without writing a single ledger query. We debit before the model call lands, refund on errors, and surface every cent on a dashboard you can hand to support.

  • Per-user wallets. Top up, debit, refund — all atomic.
  • Pre-flight checks. Block requests that would breach budget before the model spins up.
  • Tier rules. Free, Pro, Enterprise — different budgets, different rate limits.
See the billing primitives →
dashboard.assistiv.ai/end-users/user_28f3a
END USER
user_28f3a
Pro tier
BALANCE
$5.80
SPENT (30D)
$4.20
RECENT
14:02 · gpt-4o · 1.2k tok−$0.018
13:58 · claude-sonnet-4 · 890 tok−$0.012
13:51 · top-up+$10.00

Drop-in SDK

Works with whatever you're already shipping.

Next.js route handler. FastAPI endpoint. Express middleware. The SDK is a thin wrapper over fetch — no agent runtime, no orchestration framework, no opinions about your stack. You stay in your codebase. We stay out of your way.

  • Streaming first. SSE and chunked responses match your framework's idioms.
  • Per-user context. Pass user and we handle the wallet, budget, and rate limit lookup.
  • Provider failover. Anthropic 5xx → quietly retry on OpenAI. Your code never sees it.
SDK reference →
~/saas-app·npm i @assistiv/sdk
// app/api/chat/route.ts
import { assistiv } from "@assistiv/sdk";
import { auth } from "@/lib/auth";

export async function POST(req) {
  const session = await auth();
  const { messages } = await req.json();

  // pass session.userId — wallet/budget look up automatically
  return assistiv.stream({
    user: session.userId,
    model: "claude-sonnet-4",
    messages,
    fallback: ["gpt-4o", "gemini-2.5-pro"],
  });
}

AI agent-ready

Your coding agent already knows how to integrate us.

The docs are written for humans and LLMs. A curated llms.txt at the root, a live MCP docs server, Markdown-first API reference. Point Cursor, Claude Code, or v0 at our URL and the integration lands on the first prompt — no hallucinated endpoints, no stale SDK calls.

  • llms.txt + llms-full.txt. Machine-readable site map, curated for AI crawlers.
  • MCP docs server. Your agent queries the catalog live — no scraping.
  • First-prompt success. Verified against Cursor, Claude Code, Copilot, Windsurf, v0.
Read llms.txt →
$ curl·https://assistiv.ai/llms.txt
# Assistiv — AI infrastructure for SaaS
# https://assistiv.ai

> One call, any model. Per-user wallets, budgets, rate limits, tool execution.

## Quickstart
- npm install @assistiv/sdk
- Pass `user: id` — per-user wallet auto-debited
- Standard OpenAI-compatible chat/completions API

## API reference
- /docs/chat-completions  # streaming + non-streaming
- /docs/wallet            # top-up, debit, refund
- /docs/mcp               # tool catalog + OAuth
- /docs/budgets           # per-user + per-tier

## MCP docs server
mcp://docs.assistiv.ai/v1  # live catalog + examples

Tool calls + OAuth

Let your AI talk to GitHub, Slack, Stripe — without writing the OAuth.

Connect any MCP-compatible tool through a single config. Your end user clicks once to authorize their own GitHub or Slack account; the model gets a scoped, expiring token; you never store a credential. The tool call is billed to that user's wallet, capped by their budget.

  • Pre-built MCP tools. GitHub, Slack, Zendesk, Gmail, Calendar — more added monthly.
  • Per-user OAuth. Tokens scoped to the end user, not your platform.
  • Bring your own MCP. Drop in any MCP server URL and it's available to your model.
Browse the tool catalog →
live tool log·session_84a3
14:02:18github.create_issue200
14:02:14slack.send_message200
14:02:09gmail.search200
14:02:04github.list_prsretry
14:01:58calendar.find_slot200
session cost$0.0312 · debited from user_28f3a
7 min
Median time from npm install to first AI response in production.
99.95%
Inference uptime — provider failover and retries are handled by the gateway, not your code.
$0
Lines of billing code you write. Wallets, debits, refunds, and budgets are first-class primitives.
12+
MCP tools out of the box. GitHub, Slack, Stripe, Gmail, Calendar, and more — with per-user OAuth.

"We had a working AI assistant inside our app the same Tuesday we signed up. The wallet thing alone saved us a quarter of engineering work."

— Engineering lead, mid-stage SaaS · 40k MAU

Ship AI in your SaaS this afternoon.

Request access →Talk to a founder

Request access

Request access.

Tell us a bit about what you're building. Invite lands in your inbox same-day.

invite processed same-day