Codex

Codex talks to Inferam over the OpenAI Responses format.

With the CLI

inferam codex

Arguments pass through: inferam codex exec "add a test", inferam codex resume <id>. The CLI adds -c overrides for that run only. Your ~/.codex/config.toml is left alone.

By hand

Add the provider to ~/.codex/config.toml:

[model_providers.inferam]
name = "Inferam"
base_url = "https://api.inferam.ai/v1"
env_key = "INFERAM_API_KEY"
wire_api = "responses"

Create a profile in its own file, ~/.codex/inferam.config.toml. Current Codex versions refuse a [profiles.inferam] table in config.toml.

model_provider = "inferam"
model = "DeepSeek-V4.1-Flash-1017"
model_context_window = 524288

Then:

export INFERAM_API_KEY=ifr_...
codex --profile inferam

Notes

• Inferam stores no conversation state: requests must carry the full input, which Codex already does. previous_response_id is rejected.
• Hosted tools such as web search are dropped; function tools and Codex’s freeform tools (like apply_patch) work.