Codex

Codex talks to an OpenAI-compatible endpoint. sotatek-proxy forwards the path 1:1, so pointing Codex at http://127.0.0.1:3817/openai/v1 reaches Bifrost’s OpenAI-compatible route and adds local attribution headers along the way.

Bifrost currently exposes two Codex model providers on this route:

Provider	Models
fridaycodex	gpt-5.5, gpt-5.4
openai	gpt-5.5, gpt-5.4, gpt-5.5-codex, gpt-5.3-codex, and the full list

Pick one provider first, then run the matching setup command or paste the matching config.toml block below. The Bifrost key is stored in experimental_bearer_token; no separate environment variable is needed.

Note

In Codex config, model_provider = "sotatek" is the local endpoint that points at sotatek-proxy. The actual Bifrost provider is selected by the model prefix: fridaycodex/... or openai/....

Prerequisites

1. Codex CLI installed
2. sotatek-proxy installed and running on 127.0.0.1:3817
3. A Bifrost virtual key (sk-bf-xxx) — contact your admin

---

Configure with CLI

Run one command based on the provider you want Codex to use. The CLI rewrites ~/.codex/config.toml with a clean Codex config and maps the profile to the correct model id.

FridayCodex

sotatek-proxy setup -p codex -m fridayaix -k sk-bf-your-virtual-key-here

OpenAI

sotatek-proxy setup -p codex -m openai -k sk-bf-your-virtual-key-here

If the file already exists, the CLI creates a timestamped backup first. It keeps an existing experimental_bearer_token only when -k is omitted; stale providers, profiles, and model ids are removed. Older openai_api_key values are migrated into experimental_bearer_token.

Manual config.toml

Open ~/.codex/config.toml (create it if missing). If you already have top-level keys, update them to match one block below. Replace sk-bf-your-virtual-key-here with your real Bifrost virtual key.

FridayCodex

model = "fridaycodex/gpt-5.5"
model_provider = "sotatek"
model_reasoning_effort = "high"

[model_providers.sotatek]
name = "Sotatek Proxy"
base_url = "http://127.0.0.1:3817/openai/v1"
experimental_bearer_token = "sk-bf-your-virtual-key-here"
wire_api = "responses"

OpenAI

model = "openai/gpt-5.5"
model_provider = "sotatek"
model_reasoning_effort = "high"

[model_providers.sotatek]
name = "Sotatek Proxy"
base_url = "http://127.0.0.1:3817/openai/v1"
experimental_bearer_token = "sk-bf-your-virtual-key-here"
wire_api = "responses"

Verify

codex "Hello, which model are you? Reply in 1 sentence."

A successful response means Codex is routing through sotatek-proxy to Bifrost.

Available models

Keep model_provider = "sotatek" and change only the top-level model.

FridayCodex

FridayCodex has two Codex models:

Model	Notes
fridaycodex/gpt-5.5	Primary
fridaycodex/gpt-5.4	Fallback

OpenAI

OpenAI exposes the full Codex model set:

Model	Notes
openai/gpt-5.5	Primary
openai/gpt-5.4	Fallback
openai/gpt-5.5-codex	Codex primary
openai/gpt-5.3-codex	Recommended fallback
openai/gpt-5.2-codex
openai/gpt-5.1-codex
openai/gpt-5.1-codex-max
openai/gpt-5.1-codex-mini
openai/gpt-5-codex
openai/codex-mini-latest	Lightweight

Quick reference

Setting	FridayCodex	OpenAI
Config file	~/.codex/config.toml	~/.codex/config.toml
CLI profile	-m fridayaix	-m openai
Base URL	http://127.0.0.1:3817/openai/v1	http://127.0.0.1:3817/openai/v1
Provider id	sotatek	sotatek
Default model	fridaycodex/gpt-5.5	openai/gpt-5.5
Wire API	responses	responses
Key field	experimental_bearer_token	experimental_bearer_token
Key format	sk-bf-xxx	sk-bf-xxx

Troubleshooting

Issue	Solution
Authentication failed	Confirm experimental_bearer_token is set in config.toml and is a valid Bifrost virtual key.
Connection refused on 127.0.0.1:3817	sotatek-proxy is not running — sotatek-proxy status && sotatek-proxy start.
Model not found	Run setup with -m fridayaix or -m openai, or use a provider-prefixed model in manual config.
Key not picked up	Confirm the key is stored in experimental_bearer_token; no env var is required.
Wrong endpoint	base_url must end with /openai/v1 and wire_api must be responses.