Add Floe to your existing pipeline

You already have a working voice agent — STT → LLM → TTS, on Vapi, Retell, Pipecat, LiveKit, or your own loop. You don't need to rebuild it. Route the calls you already make through Floe to get one key, spend limits, and a per-call cost breakdown — without changing your agent's logic.

Pick how much you want on one ledger:

	What changes	You get
Option A — route the LLM only	One line (base URL + key)	LLM spend on one key, capped, itemized. Keep Deepgram/ElevenLabs untouched.
Option B — route every leg	Wrap each vendor call	STT + LLM + TTS on one bill, one task budget

---

Option A — route just the LLM (one line)

Floe's LLM endpoint is OpenAI-compatible, so it's a base_url + key swap. Keyless means Floe holds the upstream model key — you send only your Floe key.

Python

from openai import OpenAI

# Before:
# client = OpenAI(api_key=OPENAI_API_KEY)

# After:
client = OpenAI(
    api_key=FLOE_API_KEY,
    base_url="https://credit-api.floelabs.xyz/v1",
)
resp = client.chat.completions.create(
    model="openai/gpt-4o",          # Floe routes to the provider — no provider key
    messages=[{"role": "user", "content": "..."}],
)

TypeScript

import OpenAI from "openai";

// Before: new OpenAI({ apiKey: OPENAI_API_KEY })
const client = new OpenAI({
  apiKey: process.env.FLOE_API_KEY,
  baseURL: "https://credit-api.floelabs.xyz/v1",
});
const resp = await client.chat.completions.create({
  model: "openai/gpt-4o",
  messages: [{ role: "user", content: "..." }],
});

curl

curl -X POST https://credit-api.floelabs.xyz/v1/chat/completions \
  -H "Authorization: Bearer $FLOE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-4o","messages":[{"role":"user","content":"..."}]}'

Shortcut: npx @floelabs/cli init mints the key into your OS keychain and ends by printing exactly this base-URL swap, key filled in. See Floe CLI.

That's it — the LLM leg is now on your Floe balance, under your spend controls, and every response returns its cost in X-Floe-Payment-Amount (see Pricing).

Keep your own model key? Point at the BYOK metered proxy /v1/llm/chat/completions (model id with or without a provider/ prefix there, e.g. gpt-4o) and add X-Floe-Provider-Key: <your OpenAI/Anthropic key> — Floe meters the call and charges only a routing fee; you pay the vendor at cost. See Floe Inference.

---

Option B — route every leg (STT + LLM + TTS on one bill)

Send your STT and TTS calls through Floe's proxy too. Same request you make today, wrapped in POST /v1/proxy/fetch with your Floe key — Floe pays the vendor, so you don't manage a separate Deepgram/ElevenLabs key or balance.

# STT (Deepgram) — before: POST https://api.deepgram.com/v1/listen with your Deepgram key
curl -X POST https://credit-api.floelabs.xyz/v1/proxy/fetch \
  -H "Authorization: Bearer $FLOE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Floe-Task-Id: call-8842" \
  -d '{"url":"<DEEPGRAM_STT_ENDPOINT>","method":"POST","body":"{...audio...}"}'

# TTS (ElevenLabs) — same shape
curl -X POST https://credit-api.floelabs.xyz/v1/proxy/fetch \
  -H "Authorization: Bearer $FLOE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-Floe-Task-Id: call-8842" \
  -d '{"url":"<ELEVENLABS_TTS_ENDPOINT>","method":"POST","body":"{...text...}"}'

Tag every leg with the same X-Floe-Task-Id and one task budget caps the whole conversation — STT + LLM + TTS together. See The Voice Stack for the full costed turn.

---

Framework one-liners

Most voice frameworks let you override the LLM base URL — that's all Option A needs:

Framework	How
Vapi	Set a custom LLM provider with base URL https://credit-api.floelabs.xyz/v1 and your Floe key
Vercel AI SDK	createOpenAI({ baseURL: "https://credit-api.floelabs.xyz/v1", apiKey: FLOE_API_KEY })
Pipecat / LiveKit	Point the LLM service's base_url at https://credit-api.floelabs.xyz/v1. See the live-voice note below for the TTS and STT legs.
LangChain	ChatOpenAI(openai_api_base="https://credit-api.floelabs.xyz/v1", openai_api_key=FLOE_API_KEY)
Anything HTTP	It's OpenAI-compatible — change the host and key

Live voice (Pipecat / LiveKit): which legs land on Floe

A live LiveKit/Pipecat pipeline streams audio and wires three separate services — STT, LLM, TTS. All three route through Floe keyless, on one ledger:

• LLM — keyless base-url swap above (https://credit-api.floelabs.xyz/v1). Metered on Floe.

• TTS — Floe keyless too: point the framework's TTS service at POST /v1/audio/speech (OpenAI-compatible), or wrap a vendor in POST /v1/proxy/fetch. Metered on Floe.

• Live STT — Floe keyless too: point the framework's STT service at the streaming WebSocket wss://credit-api.floelabs.xyz/v1/audio/transcriptions/stream?model=deepgram/nova-3&encoding=linear16&sample_rate=16000&language=en with your Floe agent key. Stream PCM frames up; Floe emits the interim/final transcript events the plugin expects. Metered per audio-second on Floe (Floe fronts the Deepgram key) — no BYO key.

For the full picture, see The Voice Stack — live voice with your own stack.

---

Next

• Installation · Authentication · Pricing & cost

• Spend Controls — cap the pipeline you just connected

• The Voice Stack — the full STT→LLM→TTS turn on one key