Knowledge base

Tìm nhanh trong tài liệu Zora

Responses API For Codex CLI

Use /v1/responses when configuring Codex CLI with Zora as a custom provider. Codex should talk to the public base URL https://api.zora.io.vn/v1, read the API key from ZORA_API_KEY, and use the responses wire API.

POST/v1/responses

Responses route for Codex CLI custom providers and direct Responses-compatible API calls.

Recommended beta default

Start one tester on gpt-5.5. Production is current production, so use exact model IDs fromGET /v1/models and use exact model IDs only.

Codex Config

The official Codex config supports a custom model_provider block withbase_url, env_key, and wire_api = "responses". Put this in~/.codex/config.toml.

~/.codex/config.toml

model_provider = "zora" model = "gpt-5.5" [model_providers.zora] name = "Zora" base_url = "https://api.zora.io.vn/v1" env_key = "ZORA_API_KEY" wire_api = "responses"

Smoke Test

Set the key only in the local shell or a user-approved secret manager. Do not paste the raw key into public chat, frontend code, or a repository.

401 checklist

Copy the full, unmasked key from Dashboard API Keys. Do not use placeholder text, a masked key with* characters, an older disabled key, or a value with extra spaces or quotes.

Windows PowerShell

$env:ZORA_API_KEY = "YOUR_ZORA_API_KEY" codex exec -m gpt-5.5 "Reply with exactly: ZORA_CODEX_OK"

macOS / Linux / Git Bash

export ZORA_API_KEY="YOUR_ZORA_API_KEY" codex exec -m gpt-5.5 "Reply with exactly: ZORA_CODEX_OK"

Model Choice

Model	Use
`gpt-5.5`	Primary production smoke-test model
`gpt-5.4`	High quality chat and coding work
`gpt-5.4-mini`	Lower-latency chat and coding work
`gpt-5.3-codex`	Codex route published by Zora
`gpt-5.2-codex`	Codex route published by Zora
`gpt-5.2`	General GPT route published by Zora
`gpt-5.1-codex-max`	Higher-capability Codex route published by Zora
`gpt-5.1-codex-mini`	Lower-latency Codex route published by Zora
`gpt-5-codex`	Codex route published by Zora
`gpt-5-codex-mini`	Mini Codex route published by Zora
`gpt-5.3-codex-spark`	Spark Codex route published by Zora
`codex-auto-review`	Auto-review route published by Zora

Model IDs

Current production uses exact model IDs from GET /v1/models. Do not append client-specific suffixes or custom model-mode aliases.

Reasoning Effort

Direct Responses calls can set reasoning_effort or effort tolow, medium, high, or xhigh. Keepmodel as an exact model ID.

Responses effort

{ "model": "gpt-5.5", "reasoning_effort": "high", "input": "Review this patch and list the riskiest edge cases." }

Direct API Check

If Codex config looks correct but the CLI still fails, test the same key against the Responses route. This checks Zora authentication and model availability without running a full agent task.

curl

curl https://api.zora.io.vn/v1/responses \ -H "Authorization: Bearer YOUR_ZORA_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gpt-5.5", "reasoning_effort": "medium", "input": "Reply with exactly: ZORA_RESPONSES_OK", "max_output_tokens": 20 }'

What Zora Handles

/v1/responses for Codex CLI custom provider traffic.
/v1/chat/completions for OpenAI-compatible chat tools.
/v1/models for model discovery.
Usage accounting for requests, input tokens, output tokens, and cache token fields when they are available.
Resilient routing for chat and coding traffic, with clear retry guidance when a route is temporarily unavailable.

Troubleshooting

Symptom	Check
`401 invalid_api_key`	Confirm `ZORA_API_KEY` is set in the same shell that launches Codex and matches the full, active dashboard key.
`403 model_not_allowed`	Start with `gpt-5.5` or ask support to enable the requested model for the key.
`429` or temporarily unavailable route	Follow the retry header, reduce Codex concurrency, or move the tester to a higher RPM key.
Codex ignores Zora	Check `~/.codex/config.toml`, especially `model_provider = "zora"` and `wire_api = "responses"`.
Codex says tools are unavailable	Update to the current Zora gateway and keep `wire_api = "responses"`. Zora now bridges Codex `exec_command` tool calls for workspace tasks.

Beta test scope

For the first external user, test a tiny repository and one simple edit first. Codex can make many sequential requests on larger tasks, so start with gpt-5.5 and only move to other Zora Codex IDs after the smoke test succeeds.

For a human-oriented overview, see the Codex CLI custom provider guide.