One key. Every model. Codex and Claude Code included.

• One key, one balance. No separate accounts at Anthropic, OpenAI, DeepSeek, Moonshot and Z.ai.
• Uncensored models in your editor. Security research, red-team prompts, fiction tooling, data that trips mainstream filters. You pick the filter level, not the vendor.
• Right model per task. A frontier model for the hard refactor, a flash model for renames and test scaffolding. Same config, different id.
• Same credits as the chat app, same prices. Every response reports what it cost.

Create a key under Settings > API keys. It starts with sk-unb- and is shown once. The API is part of every paid plan: an active subscription or any credit pack unlocks it.

export UNBOTTLED_API_KEY="sk-unb-YOUR_KEY"

curl https://unbottled.ai/api/en-GLOBAL/v1/models \
  -H "Authorization: Bearer $UNBOTTLED_API_KEY"

curl https://unbottled.ai/api/en-GLOBAL/v1/chat/completions \
  -H "Authorization: Bearer $UNBOTTLED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"glm-5.3","messages":[{"role":"user","content":"Hello"}]}'

Keep the key in an environment variable, never in a file you commit. The curl call proves the key works and lists every model id your account can call, with context size, tool support and credit prices.

• Codex CLI: OpenAI Responses API.
• Claude Code: Anthropic Messages API.
• OpenCode, Continue, Aider, Cline: OpenAI chat completions.

Anything else with a base URL field works the same way: base URL, key, model id.

# OpenAI-compatible tools (chat completions or Responses API)
Base URL:  https://unbottled.ai/api/en-GLOBAL/v1
API key:   sk-unb-YOUR_KEY
Model:     glm-5.3

# Anthropic-compatible tools (Messages API)
Base URL:  https://unbottled.ai/api/en-GLOBAL
API key:   sk-unb-YOUR_KEY
Model:     glm-5.3

Codex speaks the Responses API, and so do we. The one command defines the provider inline: nothing to edit, nothing to install. For daily use, append the config once and start Codex with the profile.

UNBOTTLED_API_KEY="sk-unb-YOUR_KEY" codex \
  -c 'model_providers.unbottled={name="unbottled.ai",base_url="https://unbottled.ai/api/en-GLOBAL/v1",env_key="UNBOTTLED_API_KEY",wire_api="responses"}' \
  -c model_provider=unbottled \
  -m glm-5.3

[model_providers.unbottled]
name = "unbottled.ai"
base_url = "https://unbottled.ai/api/en-GLOBAL/v1"
env_key = "UNBOTTLED_API_KEY"
wire_api = "responses"

[profiles.unbottled]
model_provider = "unbottled"
model = "glm-5.3"

export UNBOTTLED_API_KEY="sk-unb-YOUR_KEY"
codex --profile unbottled

Switch models with -m or the model line in the profile. Keep wire_api = "responses": current Codex accepts nothing else.

Claude Code reads its gateway from environment variables. ANTHROPIC_BASE_URL is the base URL without /v1. ANTHROPIC_MODEL sets the main model. The OPUS and SONNET defaults cover the model picker and plan mode, the HAIKU default moves background jobs such as session titles to a cheap model, and CLAUDE_CODE_SUBAGENT_MODEL covers subagents. CLAUDE_CODE_MAX_CONTEXT_TOKENS tells Claude Code the real context size of a non-Claude model, so it compacts at the right point.

ANTHROPIC_BASE_URL="https://unbottled.ai/api/en-GLOBAL" \
ANTHROPIC_AUTH_TOKEN="sk-unb-YOUR_KEY" \
ANTHROPIC_MODEL="glm-5.3" \
ANTHROPIC_DEFAULT_OPUS_MODEL="glm-5.3" \
ANTHROPIC_DEFAULT_SONNET_MODEL="glm-5.3" \
ANTHROPIC_DEFAULT_HAIKU_MODEL="glm-5.3-flash" \
CLAUDE_CODE_SUBAGENT_MODEL="glm-5.3" \
CLAUDE_CODE_MAX_CONTEXT_TOKENS="1048576" \
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1" \
claude

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://unbottled.ai/api/en-GLOBAL",
    "ANTHROPIC_AUTH_TOKEN": "sk-unb-YOUR_KEY",
    "ANTHROPIC_MODEL": "glm-5.3",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.3",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.3",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5.3-flash",
    "CLAUDE_CODE_SUBAGENT_MODEL": "glm-5.3",
    "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "1048576",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

claude

Want Claude itself? Drop the model lines. Claude Code asks for its usual Claude ids and we route them to our matching Claude model, billed in credits.

Claude Code sends a big system prompt, over 30k input tokens even for a trivial task. A cheap default model keeps that affordable.

Save the config as opencode.json in your project, or as ~/.config/opencode/opencode.json for every project. The key stays in your environment. Add any id from GET /v1/models to the models map, then pick one with /models in the TUI.

UNBOTTLED_API_KEY="sk-unb-YOUR_KEY" \
OPENCODE_CONFIG_CONTENT='{"provider":{"unbottled":{"npm":"@ai-sdk/openai-compatible","options":{"baseURL":"https://unbottled.ai/api/en-GLOBAL/v1","apiKey":"{env:UNBOTTLED_API_KEY}"},"models":{"glm-5.3":{}}}},"model":"unbottled/glm-5.3"}' \
opencode

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "unbottled": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "unbottled.ai",
      "options": {
        "baseURL": "https://unbottled.ai/api/en-GLOBAL/v1",
        "apiKey": "{env:UNBOTTLED_API_KEY}"
      },
      "models": {
        "glm-5.3": { "name": "GLM-5.3" },
        "claude-sonnet-5.5": { "name": "Claude Sonnet 5.5" },
        "glm-5.3-uncensored": { "name": "GLM 5.3 Uncensored" }
      }
    }
  },
  "model": "unbottled/glm-5.3"
}

export UNBOTTLED_API_KEY="sk-unb-YOUR_KEY"
opencode

name: unbottled
version: 1.0.0
schema: v1
models:
  - name: GLM-5.3
    provider: openai
    model: glm-5.3
    apiBase: https://unbottled.ai/api/en-GLOBAL/v1
    apiKey: ${{ secrets.UNBOTTLED_API_KEY }}
    roles: [chat, edit, apply]

export UNBOTTLED_API_KEY="sk-unb-YOUR_KEY"
cn --config ./config.yaml

secrets.UNBOTTLED_API_KEY resolves from your environment. Add -p with a prompt for a headless run. The IDE extension reads ~/.continue/config.yaml: put the same model entry there.

One command, no config file. Aider wants openai/ in front of the model id and strips it before it calls us. --edit-format diff keeps edits small; without it Aider falls back to resending whole files for model ids it does not know. Name the files to edit on the command line, as usual with Aider.

OPENAI_API_BASE="https://unbottled.ai/api/en-GLOBAL/v1" \
OPENAI_API_KEY="sk-unb-YOUR_KEY" \
aider --model openai/glm-5.3 --edit-format diff --no-show-model-warnings

Two commands with the Cline CLI: save the provider once, then run cline. Base URL, key and model id are the same three values as everywhere else.

cline auth -p openai -k "sk-unb-YOUR_KEY" -b "https://unbottled.ai/api/en-GLOBAL/v1" -m glm-5.3

cline

• Copy the base URL exactly, /api/<locale>/v1 included. Without the locale you get a redirect that some clients drop. Without /api you get an HTML page instead of JSON.
• Claude Code gets the base URL without /v1. With /v1 it calls /v1/v1/messages and gets a 404.
• Use the exact id from GET /v1/models, like glm-5.3. Aider wants openai/ in front. Claude Code's built-in Claude ids are mapped for you to our newest Claude of the same family. Any other unknown id returns 404 model_not_found.
• HTTP 402 with paid_plan_required means the account has no paid plan yet. Any credit pack fixes it.
• A model without tool support returns 400 tools_not_supported once the agent sends tools. Check supports_tools in GET /v1/models.
• Each key allows 60 requests per minute. Agents fire many small calls. If you hit 429, give the agent its own key.