Coding Agents

Use Codex

Point Codex at camelStream with a named Responses provider

Codex can use a custom OpenAI Responses provider. camelStream exposes /v1/responses.


Provider	`camel_stream`
Model	`auto`
Base URL	`https://stream.camelai.com/v1`
Key	`CAMEL_API_KEY`
Context window	260K tokens guaranteed (see the fleet)

Codex ignores provider settings in project .codex/config.toml files. Put this in your user Codex home, usually ~/.codex. A separate profile keeps Stream from replacing your normal OpenAI provider.

1. Install Codex

Follow the Codex CLI installation instructions, or:

Terminal

npm install -g @openai/codex

2. Create a Stream key

Generate an API key in the Stream console. It is shown once.

Terminal

export CAMEL_API_KEY="qaml_live_..."

Keep the key in your shell environment or a secret manager. Do not put it in the TOML file.

3. Add camelStream

Add the provider to your user config:

~/.codex/config.toml

[model_providers.camel_stream]
name = "camelStream"
base_url = "https://stream.camelai.com/v1"
env_key = "CAMEL_API_KEY"
wire_api = "responses"

Then create a separate profile so Stream does not replace your normal provider:

~/.codex/camel-stream.config.toml

model = "auto"
model_provider = "camel_stream"
model_context_window = 262144
model_reasoning_effort = "high"

wire_api = "responses" is the transport for Stream's /v1/responses endpoint. Unlike Claude Code, this URL includes /v1. See Codex custom model providers.

model_reasoning_effort is sent to Stream as the Responses API reasoning.effort value. Use high for stronger reasoning, or change it to low when latency matters more. model_context_window prevents Codex from relying on its generic context fallback for the Stream model.

4. Start and verify

Terminal

cd /path/to/your/project
codex --profile camel-stream

Codex should use auto through camel_stream. If auth fails, confirm CAMEL_API_KEY is set in the same shell.

Codex may warn that metadata for auto was not found because the model is not in Codex's built-in OpenAI model catalog. The explicit context and reasoning settings above keep the important runtime values from depending on fallback defaults; the warning does not mean reasoning is disabled.

camelStream queues requests when every subscribed stream is busy. Codex should keep the HTTP connection open. See Queues & errors.

What's next?

All agents

The same key works with Hermes, Claude Code, OpenCode, and Pi.

SDKs

Use the official OpenAI and Anthropic clients.

PreviousUse Claude Code Next Use OpenCode