Accounts Pool
Accounts Pool is the CodexUse Pro feature that turns selected saved Codex accounts into one local API for another OpenAI-compatible app.
Five-minute setup
- Open Settings → Accounts Pool or use
codexuse account-pool status. - Enable the pool.
- Select the profiles you want available to the pool.
- Choose how new requests should be routed.
- Generate a local API key for the runtime that will serve the client.
CodexUse shows the current base URL in Settings, and the CLI can inspect the same setup. Keep the exact base URL and runtime-specific key together when connecting a client.
Connect a local client
Accounts Pool is for apps and tools that already speak the OpenAI Responses API shape. Typical examples: scripts, editor extensions, local agent runners, and desktop apps that want one base URL, one key, and more total headroom than a single account can provide.
curl http://127.0.0.1:PORT/v1/models \
-H "Authorization: Bearer cux_pool_..."
curl http://127.0.0.1:PORT/v1/responses \
-H "Authorization: Bearer cux_pool_..." \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.4",
"input": "Summarize what changed in this repository."
}'
Manage it from the CLI Pro
You can manage the same Accounts Pool settings from the terminal without opening the desktop app.
codexuse account-pool status
codexuse account-pool enable
codexuse account-pool profiles list
codexuse account-pool keys create
codexuse account-pool keys create --runtime=daemon
codexuse account-pool sessions list
--runtime=daemon for a headless daemon URL.
Daemon runtime
If you want the pool on a headless server or SSH session, run the CodexUse daemon and pin the port so the client keeps one stable loopback URL.
codexuse daemon start --port=3773
codexuse account-pool status --runtime=daemon --port=3773
codexuse account-pool keys create --runtime=daemon
- Requires Bun on PATH because the bundled Projects server runs on Bun.
- Use
--runtime=daemonwhen you inspect or create keys for the headless daemon. - Desktop keys will not work against the daemon URL, and daemon keys will not work against the desktop URL.
- If you also want Telegram remote control, add
--telegram-bot-tokento the same daemon start command.
Choose the model names your client will see
You choose which model IDs the pool exposes. If you want simpler setup on the other side, CodexUse can also publish aliases that keep a preferred reasoning level or speed bias behind one easier public name.
Base models
Expose the upstream model IDs you want local clients to see on /v1/models.
Reasoning aliases
Publish simpler public names that keep a preferred reasoning level behind the scenes.
Fast-mode variants
Optionally expose speed-biased variants for clients that should stay quick by default.
How routing works
New pooled sessions can start on any eligible selected profile. That gives the pool a way to spread fresh work across the accounts you selected instead of burning through one account first. After a session starts, CodexUse keeps it on that account unless a rollover is needed.
- Least used: prefer the eligible profile with the most breathing room right now.
- Round robin: spread new sessions more evenly across the selected pool.
- Session continuity: once a pooled session exists, later turns stay with it unless a rollover is needed.
What happens when one account runs out
If a selected account hits quota, a rate limit, an auth problem, or a runtime failure, Accounts Pool can move that logical session to another eligible profile and continue there. That is the practical upside of pooling: more shared headroom, smoother load distribution, and fewer hard stops for the client.
What this will not fix
Accounts Pool helps when you truly have multiple eligible accounts to route across. If two saved profiles are really the same underlying OpenAI account, they still share the same headroom. The pool can organize that setup, but it cannot invent extra quota.
Good fits
Local agent tools
Give a local tool one API while CodexUse shares quota across the accounts behind it.
One-machine automation
Keep helper scripts on the same laptop while CodexUse load-balances new work across eligible profiles.
Safer failover
Keep longer response workflows alive when one account runs out or temporarily fails.