eve-floe — spend controls for Vercel Eve agents

Give your Eve agent a budget it can't blow past. Eve agents are autonomous, durable, and run unattended — Vercel's own launch cites a lead agent that costs $5,000/yr to run. This template wires an Eve agent to Floe so every dollar it pays for an API is hard-capped server-side, with the agent tapering as it nears the limit instead of getting cut off.

Floe governs the vendor/API plane (what your agent pays — 2,000+ x402 APIs). Vercel AI Gateway governs the model plane. Together = complete agent spend control.

One key for your agent's whole vendor bill — LLM, voice, telephony, search, data — metered per call and budget-capped. Let your coding agent set it up, or wire it yourself:

New accounts get a $3 Welcome Credit (300 API credits) — no card. Set up with your AI tools → · Get a key →

A runnable Eve agent with spend control built in:

• A lead research agent that delegates web fetching to a fetcher subagent.
• The fetcher pays real x402 APIs through Floe and has its own budget cap (its own Floe key) — independent of the lead. Both the per-agent cap and your developer-wide cap are enforced on every payment.
• Each agent holds one secret (a Floe key) — never a provider or vendor key.

agent/
├── agent.ts                  # lead agent
├── instructions.md
├── connections/floe.ts       # lead's Floe budget (FLOE_AGENT_KEY)
└── subagents/fetcher/
    ├── agent.ts              # fetcher subagent
    ├── instructions.md
    └── connections/floe.ts   # fetcher's OWN capped budget (FLOE_FETCHER_KEY)

Prereq: Node.js 24+ (required by eve).

git clone https://github.com/Floe-Labs/eve-floe.git
cd eve-floe
npm install
cp .env.example .env          # then fill in the two keys

Create two agent keys — each with its own budget cap — at the Floe dashboard, set FLOE_AGENT_KEY and FLOE_FETCHER_KEY in .env, then:

npx eve dev

Ask it to research a topic. Watch the fetcher pay for sources until it nears its cap, taper, and stop at the budget — the runaway can't happen.

• Hard cap (the guarantee): the fetcher's key has a budget; over-budget x402 payments are refused server-side before they settle. A refused call means the agent is done — not that it should retry.
• Advisory (the upside): the agent reads its Floe budget advisory and tapers (fewer/cheaper fetches) as it approaches the cap, so it finishes useful work on budget instead of being cut off mid-task.
• Two tiers: each (sub)agent's own cap and your developer-wide cap are both checked on every payment.

You don't need this whole template — it's two steps to add Floe to any Eve agent. See docs/ADD-FLOE-TO-YOUR-EVE-AGENT.md.

This caps what your agent pays vendors/APIs (the x402/tool plane). Your model spend is governed by Vercel AI Gateway, not Floe — the two are complementary. We don't claim to cap model tokens.

One of three clonable templates that ship Floe spend-governance by default:

• floe-vercel-ai-starter — Vercel AI SDK agent, deploy to Vercel
• floe-crewai-starter — CrewAI crew, run on Replit
• eve-floe — Vercel Eve agent with per-subagent budgets (you are here)

MIT — see LICENSE.