⚡ Intelligent Enterprise AI Gateway · Built on LiteLLM

Route Smarter. Protect Data.
Right-Size Every LLM Request.

Flit unifies 100+ model providers behind a single OpenAI-compatible interface, automatically directs prompts to the optimal cost-performance tier, enforces enterprise DLP guardrails, and provides team-level governance.

Deploy in 5 Minutes Why Enterprise Gateway?

Gateway Stream Interactive Simulator ↗ Data Plane Trace

$ POST /v1/chat/completions · Authorization: Bearer your-virtual-key

> Virtual Model Requested: "aquila" (Auto-Classification)

> Prompt: "Format this customer address list into strict GeoJSON format"

✓ Guardrail Check: Presidio DLP scanned entities · Zero PII Exfiltrated

✓ Classifier Engine: Regex Rule matched 'Formatting/GeoJSON' · Evaluated in 0.8ms

✓ Routing Decision: Routed to aquila-fast (LiteLLM deployment: gpt-4o-mini)

★ Outcome: Optimized Tier Match · Latency: 210ms · Zero PII Exposure · Spend logged to #finance

Dynamic

Multi-Tier Cost Optimization

< 2 ms

In-Memory Classifier Overhead

MCP & A2A

Protocols & Agent Governance

Zero

Inference-Path Database Queries

Logical Architecture

The Foundation of Enterprise AI

Explore why organizations adopt Flit, what capabilities it delivers out of the box, and how its decoupled architecture guarantees extreme reliability.

Why Flit?

Uncontrolled model sprawl leads to massive budget waste, security blindspots, and vendor lock-in. Flit eliminates over-provisioning by directing work to the right model tier automatically.

Explore the Problem & ROI →

What is Flit?

A unified gateway combining smart model tiering (aquila-fast, smart, power), native MCP protocol routing, A2A autonomous agents, enterprise Presidio DLP guardrails, team-scoped virtual keys, and real-time usage analytics.

See Product Capabilities →

How It Works

Complete data plane and control plane separation. LiteLLM handles high-throughput inference; workers compile routing state in memory with zero database queries on hot paths.

Deep Dive into Architecture →

Semantic Routing

The Three Public Tiers

Developers simply request aquila or one of three transparent tiers. Flit takes care of health, failover, and exact deployment selection.

aquila-fast Low Cost & Latency

Optimized for lightweight processing, formatting, classification, extraction, and high-frequency chatbot interactions.

Target Models GPT-4o-mini, Haiku 3.5
Latency < 300 ms
Cost Profile ~ $0.15 - $0.30 / MTok
Failover Target Multi-region replica

View Fast Tier Specs

aquila-smart Enterprise Default

The versatile workhorse for multi-turn synthesis, coding, document reasoning, and nuanced natural language responses.

Target Models GPT-4o, Sonnet 3.5
Latency < 850 ms
Cost Profile ~ $2.50 - $3.00 / MTok
Failover Target Cross-provider backup

View Smart Tier Specs

aquila-power Frontier Reasoning

Reserved for complex math, formal logic proofs, deeply nested architectures, and multi-step agentic planning.

Target Models o1, DeepSeek-R1
Latency Variable (Chain of thought)
Cost Profile Frontier rates
Failover Target Dedicated fallback pool

View Power Tier Specs

Zero Client Code Rewrite

Drop-in OpenAI Compatibility

Switch your base URL and API key. Existing code, SDKs, and LangChain/LlamaIndex agents work without modification.

from openai import OpenAI

# Simply point client to your Flit gateway endpoint
client = OpenAI(
    base_url="http://gateway.yourdomain.com:4000/v1",
    api_key="flit-team-scoped-virtual-key"
)

# Use virtual model selector "aquila" for automatic classification
response = client.chat.completions.create(
    model="aquila",
    messages=[
        {"role": "user", "content": "Extract the line items and totals from this invoice..."}
    ]
)

print(f"Model selected: {response.model}")
print(response.choices[0].message.content)

curl http://localhost:4000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-virtual-key" \
  -d '{
    "model": "aquila",
    "messages": [
      {"role": "user", "content": "Summarize customer feedback trends for Q3."}
    ]
  }'

import OpenAI from 'openai';

const flit = new OpenAI({
  baseURL: 'http://localhost:4000/v1',
  apiKey: process.env.FLIT_VIRTUAL_KEY,
});

const completion = await flit.chat.completions.create({
  model: 'aquila-smart', // Explicit tier or "aquila" for auto
  messages: [{ role: 'user', content: 'Generate quarterly earnings narrative.' }],
});

console.log(completion.choices[0].message.content);

Route Smarter. Protect Data. Right-Size Every LLM Request.