Private beta · Built for Codex users

Share capacity.See every token move.

Connect your local Codex to a trusted provider machine when you need more capacity, then track exactly what you borrowed or lent. Your repository stays local. Their OAuth credentials stay put.

  • Borrowed / lent totals
  • One-time pairing
  • Per-device revoke
trusted-share — zshTLS

~/acme/checkout

$ codex

✦ Ready. What are we building?

› finish the subscription migration

Weekly capacity unavailable on this account.

Connected to trusted provider
42 ms

✦ Picking up in the same local workspace…

✓ Updated billing migration

✓ Tests passing

0OAuth tokens shared
2usage directions tracked
official Codex CLI
0600local credential files

Two sides. One local-first relay.

Borrow when blocked. Share when idle.

CodexShare keeps execution where it belongs. The consumer keeps their workspace, shell, approvals, skills, and MCP. The provider contributes only model capacity through a revocable local daemon.

01

For consumers

Continue on your own machine.

Pair once, launch the official Codex CLI, and keep every tool call inside your local environment.

$codex-connect --share "codexshare://HOST:PORT/connect/CODE"
  • Local repo, shell, MCP, and approvals
  • Saved connections with online status
  • No OpenAI login required on the consumer
02

For providers

Share with one deliberate command.

Start a time-boxed daemon, send a one-use link, then watch devices, sessions, and token flow in real time.

  • 6 hour, 24 hour, or 1 week window
  • LAN or encrypted Internet tunnel
  • Per-device credentials and instant revoke

Local usage ledger

Borrowed and lent stay separate.

Every completed Responses stream updates a private local ledger. Inspect sessions, requests, failures, traffic, and input/output tokens for any time window.

Usage overview
Last 30 days
Borrowed38,742tokens · 8 sessions
Lent142,306tokens · 21 sessions
Requests1832 failed · 98.9% complete
WithdrawableNot available yet
Borrowed Lentcodex-share --usage --days 30
Open usage console

The hosted console uses demo data today. The CLI ledger is real and remains on your machine.

Capacity moves. Control does not.

A narrower relay by design.

CodexShare is not a system-wide proxy. It accepts only the Codex Responses model surface, strips sensitive headers, and injects provider authentication at the last hop.

ConsumerOfficial Codexrepo · tools · approvals
Responses
LocalLoopback gatewayroute allowlist
TLS
ProviderShare daemonOAuth stays here

Credentials stay put

Access tokens, refresh tokens, and account identifiers never enter the pairing file or consumer environment.

Content is not logged

Product logs contain lifecycle and metering metadata, not prompt or response bodies. Sensitive fields are redacted.

Trust is explicit

The provider terminates the relay and could inspect traffic if modified. Pair only with people or teams you trust.

Deliberately narrow first release

Usage now. Money later.

P0 records usage only. The withdrawal interface is visible so the product model is clear, but there is no calculated balance, payout, wallet, or settlement behind it yet.

  1. Now
    Borrowed / lent usage ledger

    Per-session requests, failures, bytes, input tokens, and output tokens stored locally with a JSON interface.

    Available
  2. Next
    Opt-in usage sync

    A hosted read-only view of user-approved aggregates, with raw request content and OAuth credentials excluded.

    Designing
  3. Later
    Withdrawal settlement

    The interface is reserved, but balances and payouts remain disabled until pricing, authorization, reconciliation, and dispute rules are implemented.

    Not built

Why the honest version wins

No invented earnings. No pretend withdrawal.

TokenShare proves that provider/consumer framing is compelling. CodexShare starts with the measurable part: real local usage. The disabled withdrawal surface is explicit about what is still missing.

Independent projectNot affiliated with OpenAIProvider terms still apply

Questions before you connect

Read the trust boundary once.

Does the provider receive my prompts?

The provider daemon forwards model requests and therefore handles their contents in memory. CodexShare does not persist prompt or response bodies, but a modified provider could inspect them. Use a trusted provider.

Does the provider's OAuth leave their machine?

No. The daemon injects OAuth only when calling the Codex backend. The consumer gets a scoped device credential, never the OpenAI credential.

What continues to run locally?

The official Codex CLI, repository access, shell commands, file edits, approvals, MCP servers, and local skills all stay on the consumer machine.

Is this an OpenAI product?

No. CodexShare is an independent, experimental project. Users are responsible for ensuring their use complies with applicable subscription, organization, and service terms.

Private beta

Capacity should follow the team, not sit idle.

Request early access

macOS & Linux · official Codex CLI required