tokenopen-config

TokenOpen one-click client configuration plan for Claude Code, Codex, OpenClaw, Hermes, and other OpenAI-compatible developer tools.

Target host:

https://www.tokenopen.ai

Key behavior:

• Connectivity verification uses GET {host}/v1/models with Authorization: Bearer sk-xxx.
• Model and pricing discovery uses GET {host}/api/pricing.
• Claude Code uses the root host as its Anthropic base URL: {host}.
• Codex, OpenClaw, Hermes, Factory Droid, Crush, OpenCode, Oh My Pi, CodeBuddy, and most other clients use OpenAI-compatible base URL: {host}/v1.
• Codex model auto-selection prefers models from /api/pricing where:
  • tags contains LLM;
  • supported_endpoint_types is empty or includes openai;
  • model name starts with gpt-5;
  • non-small variants are preferred over mini, nano, small, lite, or tiny.
• Codex fallback model is gpt-5.5.

Compatibility

The current codebase already provides the required server-side endpoints:

• GET /v1/models
• POST /v1/messages
• POST /v1/chat/completions
• POST /v1/responses
• GET /api/pricing

Authentication compatibility:

• OpenAI-compatible clients can use Authorization: Bearer sk-xxx.
• Claude/Anthropic-compatible clients can use x-api-key: sk-xxx; maps it to Bearer auth for /v1/messages and /v1/models.

Pricing compatibility:

• /api/pricing already returns fields consumed by tokenopen:
  • model_name
  • tags
  • model_ratio
  • completion_ratio
  • cache_ratio
  • supported_endpoint_types

TokenOpen Configuration

Use this host everywhere:

https://www.tokenopen.ai

Use this root URL for Claude Code:

https://www.tokenopen.ai

Use this OpenAI-compatible URL for Codex and most other clients:

https://www.tokenopen.ai/v1

Use this pricing endpoint for model discovery:

https://www.tokenopen.ai/api/pricing

Use this model verification endpoint:

https://www.tokenopen.ai/v1/models

Manual Client Setup

Until tokenopen-config is implemented, use zcf directly.

Claude Code:

npx -y zcf init -s -T cc -f -r merge -t api_key \
  -u https://www.tokenopen.ai \
  -k sk-your-token \
  -m skip -w skip -o skip -x false

Codex:

npx -y zcf init -s -T codex -f -r merge -t api_key \
  -u https://www.tokenopen.ai/v1 \
  -k sk-your-token \
  -M gpt-5.4 \
  -m skip -w skip -o skip -x false

Claude Code environment variables:

export ANTHROPIC_BASE_URL="https://www.tokenopen.ai"
export ANTHROPIC_API_KEY="sk-your-token"
export ANTHROPIC_AUTH_TOKEN="sk-your-token"

OpenAI-compatible tools:

base_url = https://www.tokenopen.ai/v1
api_key  = sk-your-token
model    = gpt-5.4

Hermes provider shape:

providers:
  tokenopen:
    base_url: https://www.tokenopen.ai/v1
    key_env: TOKENOPEN_API_KEY
    kind: openai
model:
  default: tokenopen/gpt-5.4
  provider: tokenopen

Server Requirements

For best compatibility, TokenOpen should ensure:

• /v1/models is reachable with a valid user API token.
• /api/pricing is public or accessible without browser session auth.
• Target coding models have pricing configured.
• Target coding models are available in user groups.
• Target coding models include LLM in their tags.
• OpenAI-compatible models include openai in supported_endpoint_types, or leave the field empty.
• Claude Code models support /v1/messages.

Recommended model metadata for coding models:

{
  "tags": "LLM",
  "endpoints": {
    "openai": "/v1/chat/completions"
  }
}

tokenopen-config Implementation Plan

Required defaults:

PRIMARY_HOST = "https://www.tokenopen.ai"
FALLBACK_HOSTS = []
DEFAULT_PROVIDER_NAME = "TokenOpen"
DEFAULT_ENV_KEY = "TOKENOPEN_API_KEY"

Required CLI options:

-k, --api-key <key>
--only <client-list>
--model <model>
--base-url <url>
--skip-verify
-v, --verbose
-h, --help
-V, --version

Base URL rules:

• Claude Code: {host}
• Codex: {host}/v1
• OpenClaw: {host}/v1
• Hermes: {host}/v1
• Factory Droid: {host}/v1
• Copilot CLI: {host}/v1
• nanobot: {host}/v1
• Crush: {host}/v1
• OpenCode: {host}/v1
• Oh My Pi: {host}/v1
• CodeBuddy: {host}/v1/chat/completions

Verification:

GET https://www.tokenopen.ai/v1/models
Authorization: Bearer sk-your-token

Model discovery:

GET https://www.tokenopen.ai/api/pricing

Recommended model selection order:

1. Explicit --model.
2. Best gpt-5* LLM with OpenAI endpoint and non-small tier.
3. Any gpt-5* LLM with OpenAI endpoint.
4. Any LLM with OpenAI endpoint and non-small tier.
5. Configured fallback model.

Recommended fallback model:

gpt-5.4

Usage

Install or run directly with npx:

npx -y tokenopen-config -k sk-your-token --only claude
npx -y tokenopen-config -k sk-your-token --only codex
npx -y tokenopen-config -k sk-your-token --only openclaw
npx -y tokenopen-config -k sk-your-token --only hermes
npx -y tokenopen-config -k sk-your-token --only all

If you already installed the package globally, use this format:

tokenopen-config -k sk-your-token --only claude
tokenopen-config -k sk-your-token --only codex
tokenopen-config -k sk-your-token --only openclaw
tokenopen-config -k sk-your-token --only hermes
tokenopen-config -k sk-your-token --only all
tokenopen-config -k sk-your-token --only claude,codex
tokenopen-config -k sk-your-token --only all --model gpt-5.4

Supported --only values:

claude
codex
openclaw
hermes
all

Non-goals

• Do not modify protected project identity, package paths, or existing branding.
• Do not require changes relay routing unless a TokenOpen deployment disables /api/pricing or omits model metadata needed by client discovery.

Current Implementation

This repository now contains a working zero-dependency Node.js CLI.

Local checks:

npm run check
node ./bin/tokenopen-config.js --help

Local usage:

node ./bin/tokenopen-config.js -k sk-your-token
node ./bin/tokenopen-config.js -k sk-your-token --only claude,codex
node ./bin/tokenopen-config.js -k sk-your-token --only all --model gpt-5.4

Supported clients in this initial implementation:

• claude: uses zcf for Claude Code and writes Anthropic environment variables.
• codex: uses zcf for Codex with https://www.tokenopen.ai/v1.
• openclaw: calls local openclaw onboard with TokenOpen OpenAI-compatible settings.
• hermes: calls hermes config set when available, otherwise writes ~/.hermes/config.yaml and ~/.hermes/.env.