ProxDeck
LoginStart free

All the models. None of the paperwork.

One OpenAI-compatible endpoint for every major model, with keys and budgets handled for you.

Start freeRead the docs
Spend by org & teamSample data
Monthly hard cap
$5,950of $8,75068% · auto-stop at 100%
Core ML Eng
$3,42057%
Growth Engineering
$1,89032%
Cloud Infra & Security
$64011%
BYOK: Anthropic, AWS, AzureZero key exposure
prompt_eval.tsSample data
// ProxDeck zero-config runtime wrapper
import { ProxDeck } from '@proxdeck/sdk';
const client = new ProxDeck({
gateway: 'https://proxy.internal/v1',
model: 'enterprise:auto',
pr_attribution: '#PR-8821'
});
> Dispatching: 1,240 in / 380 out → claude-3-7-sonnet [budget ok]
Keys vaulted, never in the IDE

From the research

See the research

Fast to start. Nothing to run.

One install, no infrastructure ticket, and no ceremony between you and the first request.

Up in one sitting

Install the plugin and sign in with GitHub. Keys are provisioned to the vault for you, so there is no .env to hand around, no secret to paste and nothing to rotate by hand. No proxy daemon, no custom hardware, no infra ticket in between.

Any model on the menu

Every major provider and every current checkpoint, behind one interface. Whoever owns policy pins the allow-list once; inside it the choice is yours, and you can switch mid-sprint when a better checkpoint lands without a fresh security review.

Tooling that stays out of the way

It is a plugin, not a takeover. No proxy wrapped around your editor, no interception in your CLI, no rewritten workflow. Keep your agent, your autocomplete and your shortcuts exactly as they are.

Governed without you running it

Attribution and budget enforcement happen in the request path, so governance is not a chore you perform. No report to file, no spreadsheet to keep, and nobody chasing you for numbers at the end of the month.

Any model you want. No re-approval to switch.

The allow-list is set once by whoever owns policy, and after that the choice is yours. Swap providers mid-sprint when a new checkpoint lands. The compliance answer does not change, so neither does your velocity.

Front end teamSample data
OpenAI o3-miniallowed
Anthropic 3.7allowed
DeepSeek R1blocked
  • Every major provider behind one interface
  • Swap mid-sprint with no re-review and no ticket
  • OpenAI-compatible, so existing code keeps working
  • Keys stay vaulted, never in the IDE

One product. Three planes.

One product for keys, the IDE and the request path, from zero to governed in under 10 minutes. No proxy daemons, no custom hardware, no vendor lock-in.

Control plane

Admin portal

Vault keys. Set policy. See every request. Define model access matrices, enforce SSO, and review append-only audit trails.

  • Multi-tenant key vault
  • RBAC access matrix & model allow-lists
  • Microsoft Entra ID sync on Pro · SCIM 2.0 on Enterprise
  • Append-only audit log
Developer plane

VS Code plugin

GitHub login. No API keys in the IDE. Automatic model selection via Enterprise Auto, where the router picks an allowed model per request, and completions without ever pasting a key.

  • Single sign-on via GitHub, or Entra ID on Pro and above
  • Enterprise Auto dynamic routing
  • Zero local key leakage, no .env edits
  • Completions stream inline while you type
Data plane

Router

An OpenAI-compatible gateway with hard budgets that terminate runaway spend and granular token attribution. Overhead target: ≤2ms p50 / ≤5ms p99, on the stamp-and-lock path only.

  • Drop-in OpenAI base_url endpoint
  • Hard budget stops before bills explode
  • Every request stamped to repo, branch and PR
  • Multi-region provider failover

Start free. Get the plugin.

Free for teams of five or fewer. Bring your own keys, no token markup, cancel any time. The governance is already in the path.

Start freeSee pricing

Looking for budgets, caps, and audit? Managers landing · Back to overview