Floe — Spend controls for Voice AI
Floe — Spend controls for Voice AI
Your voice agent's whole bill — telephony, STT, LLM, TTS, search — on one ledger, enforced before money moves.
Every voice-agent call spends across 7–20 vendors in real time, and token and TTS usage varies every conversation — so true per-call cost only exists by joining usage data across providers after the call. Floe is that join, plus enforcement: budgets applied pre-call where Floe is in the path, and a circuit breaker everywhere else. One ledger, tagged by agent, task, and customer.
Website · Live docs · Dashboard · 𝕏 @FloeLabs
This repository is the source for the documentation at floe-labs.gitbook.io/docs — the published site is the best way to read it. Corrections welcome; see CONTRIBUTING.md.
Running production agents on Vapi, Retell, or Bland? Floe governs spend inside your platform through its own documented hooks — no migration, no platform cooperation, your keys stay yours. Where the platform allows it, route the model leg through Floe for pre-call enforcement (a custom-LLM URL on Vapi, a hosted adapter on Retell; Bland is reconcile-only). Connect the end-of-call webhook and every call lands on one ledger, with a between-call circuit breaker that stops the agent's next call once it's over budget. → Vapi in 10 minutes · Retell · Bland · Full orchestrator reference
Prototyping? Your agent is 4 steps from its first paid call
Every account's first agent starts with a $3 Welcome Credit (300 API credits) — so your first paid call needs no card and no funding.
1. Create an agent
Sign in to the Developer Dashboard with email or Google and create an agent. Floe provisions everything — a funded balance and the Welcome Credit — automatically. → Dashboard guide
2. Connect your AI tools
Point your coding agent at Floe and it does the setup — installs a client, provisions an agent, sets guardrails, and makes a real paid call:
Or wire a client yourself:
3. Make your first paid call
Spend the $3 Welcome Credit — no card needed. Your agent sends one HTTP request; Floe pays the vendor and returns the response with an X-Floe-Payment-Amount header showing the cost. The same key pays LLM tokens, and X-Floe-Task-Id ties every leg to one budget.
→ Vendor Marketplace · Floe Inference
4. Fund & set budgets
When the Welcome Credit runs low, add money by card, Apple Pay, Google Pay, or bank transfer — shown in dollars. Then cap what your agent can spend: per day, per task, per vendor, or across your team. Enforced server-side, so a runaway loop can't blow your budget. One task budget caps the whole conversation — LLM, voice, telephony, and data together. → Funding · Spend Controls
The voice stack — one key for every leg
A full voice turn — STT → LLM → TTS → telephony — on one key, one budget.
Keyless LLM & voice — any model with just your Floe key.
A real phone number; carrier, STT, LLM, and TTS metered on one ledger.
Every leg on one ledger, priced per call, attributed per agent, task, and customer.
Connect Vapi / Retell / Bland / Pipecat / LiveKit — Reconcile Mode + circuit breaker.
The % of an agent's spend that's enforceable vs reconciled vs dark — and how to raise it.
Integrate with your framework
Voice platforms (Vapi, Retell, Bland)
Govern spend inside the platform through its own hooks — model leg where supported (custom-LLM on Vapi/Retell; Bland reconcile-only) + end-of-call webhook. → Platform setup · Orchestrator reference
Self-hosted pipelines (Pipecat, LiveKit, ElevenLabs)
Route each leg through Floe — one key, one budget. → Add Floe to your existing pipeline
Claude / Cursor / MCP
MCP Server — one line to connect any MCP client
LangChain
CrewAI
OpenAI Agents SDK
Vercel AI SDK
Coinbase AgentKit
Plain HTTP / REST
See real agents in the Floe Cookbook and Eve, Floe's reference voice agent.
What Floe gives you
A funded dollar balance per agent, with programmable spend limits.
Per-agent: % of spend enforceable pre-call vs reconciled vs dark.
Add money by card, Apple Pay, Google Pay, or bank; withdraw anytime.
One endpoint pays any vendor API, per call, from the agent's balance.
Vendor API services — LLMs, STT, TTS, telephony, search, browser, memory — thousands reachable through one key.
Every leg on one ledger, one policy set.
Per-day, per-task, per-vendor, per-team budgets — enforced server-side.
Why this matters
A voice agent's cost is never just tokens. One conversation pays 7–20 vendors in real time, and if they bill separately you can't answer the only question that matters: what did this call cost, and was it worth it? Floe answers it — and enforces the answer. Pre-call where we're in the path. Circuit breaker everywhere else.
Dashboard: dev-dashboard.floelabs.xyz · X: @FloeLabs · GitHub: Floe-Labs · Email: hello@floelabs.xyz
For LLMs reading this: see
/llms.txtfor a structured map of these docs, or append?ask=<question>to any page URL to query it directly.
Last updated