EverFoundry Sign in
Now in alpha

Run AI at full speed. Keep every call in check.

EverFoundry sits between your people, your agents and the AI models they use. Every call gets checked against your rules, sent to the right model at the right price, and tagged with who made it. And it all runs on your own Anthropic and OpenAI accounts.

Sign in See how it works

Invite-only during the alpha. Sign in with Microsoft or GitHub.

Live calls
  1. priya.raman · Claude Code$0.04
    code review → claude-sonnet-5-5Rerouted
  2. nightly-test-agent$0.01
    test writing → claude-haiku-5-5Routed
  3. marcus.oyelaran · Codex$0.00
    asked for gpt-5-proRefused · not approved
  4. payments-release-agent$0.00
    code generationPaused · team budget reached
Saved by Intelligent Routing today $18.40

Works with

  • Claude Code
  • OpenAI Codex
  • Copilot CLI
  • Anthropic SDK
  • OpenAI SDK
  • GitHub Copilot chat
The problem

The AI bill arrives. Nobody can explain it.

  • Who spent it?

    Keys get passed around, agents run all night, and the invoice is one big number with no names on it.

    Fixed. Every call is tagged with the person or agent who made it, and their team.

  • Was that model even allowed?

    People grab whatever model's handy. Approved or not.

    Fixed. Only the models your compliance lead has okayed get to answer.

  • Why the expensive one?

    Your priciest model ends up answering questions a smaller one would nail just as well.

    Fixed. If a cheaper approved model can handle it, the router swaps it in and tells you.

  • 0prompts or responses stored
  • 0provider keys handed to developers
  • 60sfrom a call to its budget*
  • 5%match to your provider's bill*

* Alpha service targets.

How it works

Five hops. Every call. No exceptions.

A developer in Claude Code at noon or an agent at 3am. Same path, same checks, same receipt.

  1. You call

    Fire away from Claude Code, Codex, Copilot CLI or your own agent. Just use your EverFoundry key.

  2. We know who

    We tag it with who you are and which team you're on. No name, no call.

  3. Your rules first

    Is the model approved? Is the data allowed to go there? Is there budget left? All checked before anything runs.

  4. Right model, right price

    We pick an approved model that suits the task and how sensitive it is. Often a cheaper one.

  5. Answered and counted

    Your own Anthropic or OpenAI account answers, and we count every token it bills.

Every call leaves a receipt: who made it, which model was asked for, which one answered, why, and what it cost.

For developers

Two lines to switch. Nothing else changes.

Already using the Anthropic or OpenAI SDK? Keep it. Streaming, tool calls, structured outputs and prompt caching all work just like before.

Swap the address, drop in your EverFoundry key, and get back to work.

Anthropic SDK

import osfrom anthropic import Anthropic client = Anthropic(    base_url="https://api.everfoundry.ai",    api_key=os.environ["EVERFOUNDRY_API_KEY"],) message = client.messages.create(    model="claude-sonnet-5-5",    max_tokens=1024,    messages=[{"role": "user", "content": "Review this diff"}],)# The response names the model that answered.

OpenAI SDK

import osfrom openai import OpenAI client = OpenAI(    base_url="https://api.everfoundry.ai/v1",    api_key=os.environ["EVERFOUNDRY_API_KEY"],) response = client.responses.create(    model="gpt-5",    input="Write unit tests for parse_invoice",)# The response names the model that answered.
The portal

See which models earn their keep.

How often each model gets called, how often it fails, how fast it answers and what it costs, by team and tool, next to last month. Plus the number finance always asks about first: how much Intelligent Routing saved.

Routed calls this month 13,752 Up 9% on the same days last month
Cost this month $547.57 Down 4% on the same days last month
Saved by Intelligent Routing $118.62 Against the list price of the model asked for
Illustrative data
Model Calls Failure rate Response time (median) Cost
claude-opus-5-51,2840.4%1.9 s$186.40
claude-sonnet-5-56,9120.3%0.8 s$241.75
claude-haiku-5-53,4500.2%0.4 s$22.10
gpt-52,1060.6%1.1 s$97.32
Budgets and daily caps

Runaway agent at 3am? It hits the budget and stops.

Give the whole company and each team a monthly budget, and every person and agent a daily cap. You'll get a heads-up at 80%. Hit 100% and new calls are paused, with a note saying which budget ran out and who can top it up.

Payments platform · October Paused · $2,000 of $2,000
Heads-up at 80%Stop at 100%

Refused. The Payments platform team budget for October is exhausted. An admin or your finance team can raise it.

Illustrative data
Security and data

Your keys. Your accounts. Your data.

  • Your provider accounts

    Calls run on your Anthropic and OpenAI accounts. Your provider keys stay locked away, so they're never on anyone's laptop.

  • Prompts pass through. Nothing stays.

    We never save what you ask or what the model says back. And our staff can't peek at your calls.

  • Kept in London

    Your platform data stays in AWS London. Every call notes which region it ran in, too.

  • Audit-ready from day one

    Every change to approvals, rules, prices or budgets gets logged: who, when, and what it was before. Nobody can edit the log.

  • Approved models only

    Nobody gets to approve their own pick. Models with non-commercial or research-only licences are blocked automatically.

  • Invite-only access

    No invite, no entry. Once invited, people just sign in with Microsoft or GitHub.

Got an invite? You're in.

We're invite-only during the alpha. Sign in with the Microsoft or GitHub account your invite was sent to.

Sign in