Check out the newest way to compare different models for a task/agent harness: AutoEvals
Connect an agent

Coding Agents

Codex

Route OpenAI Codex through the Inference.net gateway with inf codex.

inf codex routes Codex through https://api.inference.net/v1 using the OpenAI Responses surface. Codex resolves provider credentials through an env var name, so the key lives in $CODEX_HOME/.env, not inline in the config.

Config files

inf writes fields in ~/.codex/config.toml (TOML) and $CODEX_HOME/.env. Both honor CODEX_HOME. inf owns fields in both files, not the whole files.

inf codex on

inf codex on
FlagRequiredDescriptionDefault
--model <alias>NoModel alias written as the root modelkimi-k3-fast

See Model mapping for supported aliases and limits.

What gets written

TOML owned fields in config.toml:

PathValue
modelselected alias
model_providerinference
[model_providers.inference].nameInference.net
[model_providers.inference].base_urlhttps://api.inference.net/v1
[model_providers.inference].env_keyINFERENCE_API_KEY
[model_providers.inference].wire_apiresponses

.env owned line:

PathValue
INFERENCE_API_KEYgateway key (secret)
model = "kimi-k3-fast"
model_provider = "inference"

[model_providers.inference]
name = "Inference.net"
base_url = "https://api.inference.net/v1"
env_key = "INFERENCE_API_KEY"
wire_api = "responses"

inf codex status

inf codex status

inf codex off

inf codex off

off restores unchanged owned fields. It keeps fields that you edited after on.

Examples

Select the Fast-tier alias:

inf codex on --model glm-5.2-fast

Restart Codex after on or off. See Troubleshooting for drift, recovery, and key revocation.

On this page