Codex CLI with AICredits

Route Codex CLI (@openai/codex) through AICredits to pay in INR, set per-key budget caps, and see every request in your usage dashboard.

Prerequisites

• An AICredits account with credits — top up here
• Codex CLI installed (npm install -g @openai/codex)

Setup

1. Create an API Key

Go to Dashboard → API Keys and create a new key. Optionally set a budget limit (e.g. ₹500) to cap what a single Codex session can spend.

2. Add a custom provider to ~/.codex/config.toml

Codex CLI talks to providers through a model_providers table in its config file. Add an aicredits provider and point Codex at it:

[model_providers.aicredits]
name = "AICredits"
base_url = "https://api.aicredits.in/v1"
env_key = "AICREDITS_API_KEY"
wire_api = "responses"

model_provider = "aicredits"
model = "gpt-4o-mini"

Then export the key under the exact env var name you set as env_key:

export AICREDITS_API_KEY=sk-your-aicredits-key-here

source ~/.zshrc

3. Verify it works

codex exec -c model_provider=aicredits -c model=gpt-4o-mini "say hi"

The session header should show provider: aicredits. Check Dashboard → Usage afterward — the request and INR cost should appear within a few seconds.

Switch models per session

Override model for one run without editing the config file:

codex exec -c model_provider=aicredits -c model=anthropic/claude-sonnet-4.6 "refactor this function"

Any AICredits-routed model works here — see the model catalog for the current list and pricing.

Non-OpenAI Models with Responses

Codex CLI uses the Responses API for custom providers. AICredits accepts POST https://api.aicredits.in/v1/responses: OpenAI models are forwarded to OpenAI's Responses API, while non-OpenAI models are converted to Chat Completions internally and routed through the normal AICredits pipeline.

Use a non-OpenAI model ID directly in config.toml:

[model_providers.aicredits]
name = "AICredits"
base_url = "https://api.aicredits.in/v1"
env_key = "AICREDITS_API_KEY"
wire_api = "responses"

model_provider = "aicredits"
model = "deepseek/deepseek-v4-flash"

Or override it for one run:

codex exec -c model_provider=aicredits -c model=deepseek/deepseek-v4-flash "summarize this diff"

Known limitations for non-OpenAI models:

• Responses-only features that do not have a Chat Completions equivalent may not apply.
• instructions, input, max_output_tokens, reasoning.effort, function tools, and common sampling fields are converted to their Chat Completions equivalents.
• Tool-calling support depends on the selected model.
• For best compatibility, start with simple text prompts and add tools or structured output after a basic request appears in Dashboard → Usage.

Making aicredits the default provider

If you want every codex invocation to use AICredits without passing -c model_provider=aicredits each time, set it directly in the config file (as shown in step 2) rather than overriding per-command. To switch back to OpenAI directly for a single session:

codex exec -c model_provider=openai "..."

Troubleshooting

401 Unauthorized: Invalid API Key The env var named in model_providers.aicredits.env_key is missing, misspelled, or holds an invalid key. Confirm it's set (echo $AICREDITS_API_KEY) and that it's an AICredits key from Dashboard → API Keys, not an OpenAI key.

402 Payment Required Wallet balance is zero or the key has hit its budget cap. Add credits at Dashboard → Billing or raise the key's budget limit.

Error loading config.toml: wire_api = "chat" is no longer supported Your Codex CLI version rejects wire_api = "chat" for custom providers. Change it to wire_api = "responses" — see step 2.

unknown configuration field Codex CLI validates config.toml strictly when --strict-config is passed (and increasingly by default on newer versions). Check for typos in the model_providers.aicredits block, and make sure model_providers is a top-level table, not nested under something else.