Standard Compute

All fixes · ChatGPT login in Hermes Agent / OpenCode · 429 · Quota & billing

“The usage limit has been reached” (429) in Hermes Agent and OpenCode: Fix

HTTP 429: The usage limit has been reached

Your ChatGPT plan's Codex allowance is used up: the agent is signed in with your ChatGPT account, and OpenAI's Codex backend sends this 429 once the allowance is spent. Wait for the reset shown at chatgpt.com/codex/settings/usage, buy Codex credits, or set a fallback provider so the agent continues on a different key.

What causes it

  • The response carries the error type usage_limit_reached with your plan type and the reset time. Codex CLI reports the same condition as “You've hit your usage limit”.
  • Hermes's openai-codex provider and OpenCode's ChatGPT Plus/Pro login both send requests to the Codex backend, so they spend the same allowance as Codex CLI, the IDE extension and the Codex app.
  • Plus has a five-hour window and a weekly limit. Pro plans have no five-hour window but still a weekly allowance.
  • In OpenCode, a stale login to an old free ChatGPT account instead of the paid one has produced this error (OpenCode issue #17614).

How to fix it

  1. Check which limit you hit and when it resets at chatgpt.com/codex/settings/usage, or run /status in Codex CLI.
  2. Hermes: add a fallback in ~/.hermes/config.yaml under fallback_providers (or run hermes fallback add). A model pinned for the session bypasses the fallback list (issue #117806), so unpin it.
  3. OpenCode has no automatic fallback yet: switch model with /models, or add another provider with /connect and choose it when the plan runs dry.
  4. Buy Codex credits or upgrade the plan if you want to stay on the ChatGPT login.
  5. Point the agent at an OpenAI-compatible endpoint: in Hermes use provider: custom with base_url and key_env; in OpenCode add a provider using @ai-sdk/openai-compatible with options.baseURL.

Running an agent? A persistent agent on a ChatGPT login shares one allowance with every other Codex surface you use. Running Codex CLI during the day and Hermes at night drains the same window from both ends.

Sources: openai/codex: usage_limit_reached mapping; Hermes test fixture for this 429; Hermes docs: fallback providers; OpenCode docs: providers; OpenCode issue #17614

Keep working on a fixed monthly plan

Both agents accept a custom OpenAI-compatible endpoint, so Standard Compute can be the fallback or the main provider: base URL https://api.stdcmpt.com/v1, model standardcompute. Plans cost a fixed $19 to $249 a month for a set compute budget, with no 5-hour or weekly windows. Each key gets 300 to 2,000 requests a minute by plan, and at 100% of the budget new requests return HTTP 402 until renewal, with no overage charge.

Get your API key How it connects

Plans from $19 a month with $20 of compute. Eligible new accounts can try $0.25 of compute first, no card. Trial terms

FAQ

Is this a Hermes or OpenCode rate limit?

Neither. The 429 body comes from OpenAI's Codex backend and says your ChatGPT plan's Codex allowance is spent.

How long until it works again?

The response includes the reset time. On Plus the five-hour window reopens with your next message after it closes; the weekly allowance resets seven days after it started. OpenAI support does not grant manual resets, but a banked reset, if your account has one, restores both windows from /usage in Codex CLI.

Will a fallback provider cost me anything?

Only what that provider charges. A pay-per-token key bills each request; a flat plan bills its monthly price and stops at its budget.

Related errors