TemprTempr
Sign in Start free
For engineering teams using more than one AI tool

Let your team use any AI. Keep control of all of it.

Run Claude Code on your Claude plan, Codex on ChatGPT, and Cursor, your own apps or any tool on your API keys, with one set of budgets, guardrails and logs over all of it.

Start free Plan a team rollout

Free for 10,000 requests a month. No card, and your developers keep their tools.

Works with the tools your team already uses
Claude CodeCodexCursorClineZedVS CodeJetBrainsClaudeChatGPTyour apps
acme-robotics · live requests sample data
Tool and modelPaid byWhat Tempr did
4,812 requests todayplans $1,284 valuekeys $61.20 billed23 stopped

Every request, from every tool, in one log. Who paid, and what Tempr did.

The problem

Your team already runs five AI tools. Can you answer four questions about them?

What did we spend on AI last month, and on what?

TodaySix invoices, two plans and some expense reports.
With TemprSpend by team, person, tool and model, including what each plan's requests would have cost at API prices.

Where are our API keys?

TodayIn .env files, CI secrets and shell history.
With TemprProvider keys stay in Tempr. People and apps get virtual keys you can limit, rotate or revoke in one place.

Did anything sensitive go into a prompt?

TodayNo way to know.
With TemprSecrets and personal data are flagged, masked or blocked, in what people send and in what the model answers.

Which tools can our agents call?

TodayWhatever anyone installed, signed in as whoever set it up.
With TemprTool Packs decide which MCP tools each person, group or key gets. Every call is checked, and each person signs in as themselves.
What you get

Three things every team asks for.

Any tool, any plan, any key

Keep the plans you already pay for. Govern them like everything else.

Claude Code stays on your Claude plan and Codex on your ChatGPT plan, while API keys cover your apps and everything else. Tempr records which plan or key paid for every request.

  • Claude Pro, Max, Team and Enterprise; ChatGPT Plus, Pro, Business and Enterprise
  • 30 coding plans as providers, from Z.AI and Kimi to MiniMax and Alibaba Cloud
  • When a plan is used up, the request moves to the model you chose instead of failing
How plans work with Tempr →
Coding plans · this month sample data
Claude Max4 members · plan login1,284 req$612 value
ChatGPT Business3 members · Codex902 req$388 value
Z.AI coding planzai-coding/glm-5.32,410 req$96 value
OpenAI keycheckout-api, support-bot18,220 req$41.30 billed
When Claude Max is used up → zai-coding/glm-5.3 on the Z.AI plan, so nobody's session stops.
Spend you can cap

Budgets at every level, down to a single agent run.

Every budget on a request's path applies, from the organization to the key, and a run that would cost too much stops before its next model call. Limits block; they don't send you a surprise invoice.

  • Monthly budgets for the organization, teams, groups, members and keys
  • Budget rules by hour, day, week or month, on cost, tokens or requests
  • Budgets and rate limits for your app's own end users
How spend controls work →
Budgets on one request's path sample data
Organizationacme-robotics
$920 / $2,000
Teambackend
$378 / $600
Memberpriya@
$57 / $150
Keyci-review-bot
$18 / $20 day
Agent run#4821
stopped at $2
day cost 20 blockweek requests 5000 warnmonth tokens 40M block
Data and tools stay governed

Secrets stay out of prompts. Tools stay in bounds.

Tempr checks what people send and what models answer, in the IDE, the terminal, agent runs and your apps, and it checks every MCP tool call against the tools each person is allowed.

  • Flag, mask or block secrets and personal data, with your own patterns too
  • Tool Packs: the MCP tools each person, group or key may call
  • Per-member sign-in to MCP servers, an audit log, and SSO
  • Popular remote MCP servers, such as Linear, Notion and Sentry, with sign-in filled in
How guardrails work →
Guardrail test sample data
Deploy failed. Here's the env: AWS_KEY=AKIAIOSFODNN7EXAMPLE, owner dana@acme.io
↓ sent to the model as
Deploy failed. Here's the env: AWS_KEY=[aws-key], owner [email]
github · create_pull_requestin Tool Pack
linear · create_issuein Tool Pack
jira · delete_projectrefused
How it works

One setting per tool. Rules set once. Everything in one place.

01

Point each tool at Tempr

Developers keep their tools and their sign-ins. They change one setting and add a Tempr key.

// ~/.claude/settings.json: keeps your Claude plan
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.temprhq.io",
    "ANTHROPIC_CUSTOM_HEADERS": "x-tempr-key: tvk_..."
  }
}
02

Set the rules once

In the Portal, for the whole organization or one team, group, person or key: which models and tools, how much to spend, and what never leaves the building.

03

See everything

Every request from every tool in one log: who sent it, which plan or key paid, what it cost, and what Tempr changed or stopped. Agent runs on Tempr are traced call by call, and on Pro and up you can ask Claude Code, Codex or ChatGPT about any of it.

Tempr's own agents

Want an agent too? Ours runs on the same rules.

A coding agent on your own keys and plans, in the IDE, the terminal and its own app. It follows your organization's published rules and skills, and it stays inside the same budgets and guardrails as every other tool.

Tempr Chat in VS Code proposing a change as a diff
In your IDE

Tempr Chat

Visual Studio, VS Code and JetBrains IDEs: agent, chat, completion and MCP tools.
The Tempr CLI asking for approval before an edit
In your terminal

Tempr CLI

Interactive, piped into scripts, in CI, and inside editors like Zed through ACP.
The Tempr Code desktop app
On your desktop

Tempr Code Soon

Parallel chats on their own copies of your repo, plan mode, rewind, and any model, local ones too.
Getting started

Try it on one tool today. Roll it out when you're ready.

Start on your own

Free

$0 · no card

  • 10,000 requests a month across every tool and app
  • Your own provider keys and plans
  • Request logs, virtual keys, and guardrails that flag or block
  • Upgrade in the Portal whenever you outgrow it
Start free
Roll it out to a team

From $49 a month

Gateway Pro with one seat included, then $35 a seat. Scale and Enterprise for larger teams.

  • Masking, your own data rules, answer scanning and budget rules
  • Teams, groups, Tool Packs and plan logins for your organization
  • Tempr's agents for every seat, on the same keys and rules
  • SSO and a rollout plan with us on Teams and Enterprise
Plan a team rolloutCompare setups
See full pricing →
Your keys, billed by your providersTempr never resells tokens or takes a cut. Your Tempr plan pays for the product, not the model.
Limits that holdBudgets and caps block. Nothing past a cap is ever charged.
SSO and an audit logSign in with your identity provider, and see who changed what.
Your customers stay anonymousEnd-user ids are hashed before any provider sees them.
Logs on your termsKept 7, 30 or 90 days by plan, and can be streamed to Langfuse, Opik and other tools.
Questions

Before you point a tool at us.

Do my developers have to change tools?

No. Claude Code, Codex, Cursor, Cline, Zed and your own apps keep working the way they do. Each one gets a Tempr base URL and a Tempr key, and developers keep their own sign-ins.

How are Claude and ChatGPT plans billed?

The plan keeps paying, exactly as before. Tempr relays the request, applies your rules, and records it with what it would have cost at API prices, so you can see what each plan is worth.

Do you mark up model prices?

No. You pay your providers directly on your own keys and plans. Your Tempr plan pays for the gateway, the Portal and the agents, priced by requests and seats, never by tokens.

What if Tempr has a problem?

Every tool goes back to its provider by changing one setting, because your keys and plans are still yours. Fallback chains also move a request to another model or provider when one fails.