Every agent. Every API. One place to see it all.

Your AI agents have
full access to everything.
That's the problem.

Your agents touch Stripe, OpenAI, and Anthropic. When something goes wrong, you're logging into three dashboards to figure out what happened. Fencepost gives you one place to see everything, set boundaries, and freeze any agent instantly.

Try it free — start in 60 seconds

Why Fencepost exists

Built by someone who hit this wall

Shared API keys and zero visibility burned us in production — the same pattern teams describe in calls. Fencepost is the control layer we wished existed.

Follow along on X →
See the problem ↓
The pattern nobody plans for
"You deploy one agent with a shared API key. It works. You deploy a second the same way. By the tenth, every agent has full access, identical credentials, and zero visibility. No one made a bad decision. They just made the same small one ten times."

— a pattern we hear constantly from teams shipping agents

You can't tell which agent made a request. You can't stop one without stopping all of them. You can't see what's happening until something breaks. Fencepost fixes that — a control layer you add in two lines of code.

How it works

Use the official OpenAI , Anthropic or OpenRouter SDKs.

Same client libraries you already trust—routed through Fencepost for keys, policies, and observability.

TypeScript
import OpenAI from "openai";

const openai = new OpenAI({
  apiKey: "sk-proj-unused",
  baseURL: "https://getfencepost.xyz/openai/v1",
  defaultHeaders: {
    "X-Fencepost-Token": process.env.FENCEPOST_TOKEN,
  },
});

const response = await openai.chat.completions.create({
  model: "gpt-4o-mini", // same SDK—point it at Fencepost
  messages: [{ role: "user", content: "Hello!" }],
});
Try for free → OpenAI SDK walkthrough
Features

Every agent gets an identity.
You get control.

01 / IDENTITY

Always know which agent did what

Stop sharing one API key across bots. Each agent gets its own fpt_ token, so you can trace every request back to a name — not a guess.

02 / BOUNDARIES

Let bots read data without owning your account

Scope at the endpoint: your payment helper can list charges but never issue refunds. Rules you set, not hope for.

03 / VISIBILITY & SPEND

Know why your bill spiked — before morning

Full request history per agent. Pair it with caps, alerts, and auto-freeze when spend hits the limit so a loop cannot rack up hundreds overnight.

04 / CONTROL & SETUP

Stop a runaway agent in one click

Freeze a single agent from the dashboard; everything else keeps running. Two lines in your existing OpenAI or Anthropic client — ship new agents without losing sleep.

Built for developers who
want to move fast and stay in control.

Solo devs & indie hackers

  • → You have 3 agents sharing one key — Fencepost gives each one its own lane
  • → You deserve proper security tooling — without a weekend lost to config. Here it is.
  • → Two lines of code. No Docker, no certs, no weekend lost to config.
  • → 7-day free trial with every feature — then the $5/month limited launch deal if you stay.

Small startup teams

  • → See which agent made that expensive API call — instantly
  • → New team member? Give them a scoped token, not your master key.
  • → Someone leaves? Revoke access in one click.
  • → Audit log ready to export when you need it
Pricing

Simple. Predictable.

One plan. No seat fees. 7-day free trial, then the $5/month limited launch deal.

Everything included
Limited launch deal
$5
/ month
Current launch price after your 7-day free trial.
  • · Full access to every feature during the trial
  • · Unlimited agents
  • · 90-day audit log
  • · Freeze & scope controls
  • · Per-agent budgets and alerts
  • · OpenAI & Anthropic (other providers coming soon)
Try it free →

After 7 days, the launch deal is $5/month. Cancel anytime from your dashboard.

FAQ

Questions we hear a lot

Quick answers before you wire up your first agent.

What is Fencepost, in one sentence?
It is a control layer in front of your LLM and API traffic: per-agent tokens, policies, visibility, and a one-click freeze — so shared keys and blind spots stop being the default.
Do I have to abandon the official OpenAI or Anthropic SDKs?
No. You keep the same client libraries and change the base URL plus a small header so traffic routes through Fencepost. That is the whole integration story on the code side.
Where do my real provider API keys live?
Your OpenAI and Anthropic keys stay in Fencepost under your account, encrypted at rest (AES-256). Each agent gets its own fpt_ token for runtime — so you never paste a master key into every bot or workflow.
What does Fencepost see when my agents make requests?
Because requests pass through the proxy, Fencepost can record what you need for an audit trail and spend visibility — the same reason you can tell which agent called which endpoint. Treat it like any other security middle tier: route production traffic only if you are comfortable with that visibility tradeoff.
Will this add noticeable latency?
Any proxy adds a network hop. In practice the overhead is usually small compared to model inference time, but if you are shaving milliseconds for a latency-critical path, measure it in your region and workload.
Which providers are supported today?
OpenAI and Anthropic are supported now, with the same official SDKs you already use. Additional providers are on the roadmap — the landing page will stay current as they ship.
Can I stop one agent without taking down the rest?
Yes. Freeze or revoke a single agent token from the dashboard; other agents keep their own credentials and keep running. That is the point of giving each agent its own lane.
How does pricing work after the free trial?
You get seven days with full feature access. After that it is the limited launch deal at $5 per month — unlimited agents, no seat math. Cancel any time from your billing page; you are not locked into a long contract.

Still stuck? Start free — setup takes about a minute.

Your agents are running.
Do you know what
they can access?

Add a control layer before something breaks. Set up takes less than a minute.

Start in 60 seconds →
AES-256 encryption at rest · No vendor lock-in · Your keys stay yours