Codex Pooler

What Codex Pooler does

Clients send Codex backend or OpenAI-compatible requests using a Pool key. Codex Pooler chooses an eligible upstream account based on model support, quota evidence, limits, session continuity, routing policy and health, so the key stays the same while accounts change behind it. It works with a single upstream account for credential isolation and usage visibility, and routes across several accounts when you add more.

Operators manage Pools, accounts, API keys, routing, request accounting, audit logs and health from a browser. The README says it does not store prompts, files, audio, images, bearer tokens or raw Codex secrets. Other functions include per-key request limits and daily or weekly usage allowances, alerts through the dashboard, email or webhooks, detection of provider model downgrades, an optional per-key usage dashboard and an allowlist of approved networks. For a local instance the base URL for other harnesses and SDKs is http://localhost:4000/v1, and Docker Compose or Kubernetes are the deployment routes.

Key features

• Stable Pool API keys that keep working when upstream Codex accounts are added or replaced
• Account selection by model support, quota, limits, session continuity and health
• Per-key request limits and daily or weekly usage allowances
• Request accounting and audit logs that avoid storing prompts, replies or files
• Alerts for low capacity and account problems by dashboard, email or webhook
• Docker Compose and Kubernetes deployment

When to use it

• Sharing a pool of Codex accounts across a team with one key per person or app
• Pointing OpenAI-compatible SDK apps and coding agents at a single internal endpoint
• Watching usage per key and spotting provider model downgrades

Who it is for: Teams and operators who run several coding agents or apps against Codex accounts and want one self-hosted control point.

How it fits with Hermes Agent

Codex Pooler is a general gateway for many harnesses; the repository lists hermes-agent among its topics and offers an OpenAI-compatible /v1 base URL for harnesses and SDKs.

Requirements: At least one upstream Codex account, and Docker Compose or Kubernetes to host the gateway

FAQ