EverFoundryEverything between your team and the model. Handled.
Five pieces, one service, all running on your own Anthropic and OpenAI accounts. Here's what each one does for you.
Sign inSpeaks your tools' language.
Works with the Anthropic and OpenAI APIs your tools already use. Change the address, keep your code.
- Streaming, tool calls, structured outputs, prompt caching and long sessions all work just like before.
- Everyone gets their own EverFoundry key, and so does every agent. Each one lasts up to 30 days.
- On managed devices there's no sneaking around it. Claude Code, Codex and SDK agents can't call a provider directly.
- Adds less than a second before the answer starts flowing.*
| Coverage | Tools |
|---|---|
| Always on | Claude Code |
| Always on, via your network | OpenAI Codex, agents built on the OpenAI or Anthropic SDKs, Copilot CLI |
| Opt-in | GitHub Copilot chat and agent mode with your own key, as an experiment |
* Alpha service target.
The right model for each job.
Your rules run first. Then we match the task, and how sensitive it is, to the approved models allowed to handle it. The default routing rules come ready to go, so you're routing from day one.
- Everyday
Internal stuff, no personal data.
- Confidential
Client code or anything confidential.
- Regulated
Personal or sensitive data, or regulated client work.
| Task type | Everyday | Confidential | Regulated |
|---|---|---|---|
| Code completion | haiku-5-5, sonnet-5-5 | haiku-5-5, sonnet-5-5 | sonnet-5-5 |
| Code generation | sonnet-5-5, opus-5-5 | sonnet-5-5, opus-5-5 | opus-5-5 |
| Code review | sonnet-5-5 | sonnet-5-5, opus-5-5 | opus-5-5 |
| Test writing | haiku-5-5, sonnet-5-5 | sonnet-5-5 | sonnet-5-5 |
| Documentation | haiku-5-5, gpt-5-mini | sonnet-5-5, gpt-5 | sonnet-5-5 |
| Requirements and design analysis | sonnet-5-5, gpt-5 | opus-5-5, gpt-5 | opus-5-5 |
| General questions | haiku-5-5, gpt-5-mini | sonnet-5-5, gpt-5 | sonnet-5-5 |
Illustrative data. Model names shortened.
-
Cheaper when it's good enough
If a cheaper approved model can do the job, it gets the call. You'll always see which model answered.
-
A safe fallback, or a clear no
Model down? A backup steps in, but only if it's approved and at least as capable. If not, you get a straight answer about why.
-
Steady for the whole conversation
We pick a model once per conversation and stick with it, so answers don't jump around. Claude Code always gets Claude.
-
Every decision explained
Every routing decision comes with a reason. Change a team's sensitivity level and it kicks in straight away, with who and when on record.
Only the models you've signed off.
We keep the facts on every model: who made it, its licence, where it comes from, what it can do and what it costs. Your compliance lead picks the exact versions your teams can use.
The life of a model
- Proposed
- Under review
- Approved
- Suspended
- Excluded
- Retired
Only approved models can answer a call.
- Approve an exact version. When a new one lands, it's a one-click approval.
- Nobody gets to approve a model they suggested.
- Non-commercial and research-only licences are blocked automatically.
- Rule out models by where they're made or where they're hosted.
A no that tells you what to use instead
Refused. gpt-5-pro is not approved for your workspace.
Approved alternatives: gpt-5, claude-opus-5-5
Know what every call cost. To the token.
Every line of every bill, priced from the providers' list prices, tracked to a fraction of a cent and shown in pounds or dollars.
Units billed × list price, for every component, added up.
- input
- output
- prompt caching
- reasoning
- long-context and batch rates
- tool charges
Worked out in the provider's own currency, down to a fraction of a cent. We only round the totals you see on screen.
-
Never stuck on exchange rates
Your finance team sets the rates. Missed one? We use the European Central Bank rate instead, so a call never gets blocked over a missing rate.
-
Budgets with teeth
Monthly budgets for the company and each team, plus a daily cap for every person and agent. You'll hear about it at 80% and 100%. When a budget runs out, new calls are paused.
-
Matched to the bill
Drop in each provider's monthly cost export and we'll check it against our numbers, to within 5%.* Discounts and credits show up as a monthly adjustment.
-
Exports finance can use
A monthly spreadsheet of cost by team, and every call sliceable by person, agent, tool, model, provider, task, day or month.
* Alpha service target.
One portal. The right view for every role.
Everyone sees what their role needs, and nothing more. Figures are never more than 15 minutes old.*
Developer
Your own usage and cost, and your EverFoundry key.
Team lead
Each person's usage, the team total, and budget alerts.
Admin
Invites, teams, roles, sensitivity levels, budgets and daily caps.
Finance owner
Exchange rates, budgets, cost exports and reconciliation.
Compliance Owner
Model approvals, and tracing any questioned output.
- Got a questionable answer? Trace it back to the person, session, model, version, reason and cost.
- An approval register your auditors can pull into a spreadsheet.
- A log of every change to approvals, rules, prices, budgets and daily caps. Nobody can edit or delete it.
- Light or dark, and your choice follows you to every device.
* Alpha service target.
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.