AI API, một endpoint
cho mọi model.

You are configuring Claude Code to use MotAPI (motapis.com) as its API endpoint.

Use these EXACT values. Do not invent alternatives and do not use any other host:
  ANTHROPIC_BASE_URL   = https://motapis.com          (no /v1 -- the SDK appends it)
  ANTHROPIC_AUTH_TOKEN = sk-mot-THAY_BANG_KHOA_CUA_BAN
  ANTHROPIC_API_KEY    = ""                            (empty string, see step 3)
  model                = claude-opus-4-8

Steps:
1. If ~/.claude/settings.json exists, copy it to ~/.claude/settings.json.bak first.
   Show me the diff before you write anything.
2. Create or merge ~/.claude/settings.json so it has an "env" object containing the three
   values above. Keep every other key already in that file exactly as it is.
3. ANTHROPIC_API_KEY must be PRESENT and set to the empty string "" -- not omitted. If it
   is merely absent, Claude Code falls back to authenticating against real Anthropic
   whenever this machine has a real Anthropic key in its environment, and the requests go
   to the wrong provider without any error.
4. Verify: run  claude --print "reply with exactly: pong"  and show me the output.

IMPORTANT
- Never print, echo, or commit the API key, and never write it into a file inside a git
  repository.
- Do not modify any other provider, profile, or key in that file.
- If a step fails, stop and tell me. Do not try a different host or a different auth method.

You are configuring OpenAI Codex CLI to use MotAPI (motapis.com).

Use these EXACT values:
  provider id = motapis                      (no hyphen -- Codex fails to resolve provider
                                              ids that contain "-")
  base_url    = https://motapis.com/v1
  env var     = MOTAPIS_API_KEY
  model       = gpt-5.6-terra                (Claude works too, e.g. claude-sonnet-5)

Steps:
1. Back up ~/.codex/config.toml if it exists. Show me the diff before you write.
2. If that file contains a block marked "BEGIN codex-router-provider-managed", STOP and
   tell me. codex-router rewrites that block and sets requires_openai_auth = true, which
   makes Codex send OpenAI credentials instead of the MotAPI key -- the request then never
   reaches MotAPI and fails with 401 api_key_not_supported. Either disable codex-router
   first, or keep using it and pick a custom/motapis-* model inside it instead.
3. Add or update this block, keeping every other setting in the file unchanged:

model_provider = "motapis"
model = "gpt-5.6-terra"

[model_providers.motapis]
name = "motapis"
base_url = "https://motapis.com/v1"
requires_openai_auth = false
env_key = "MOTAPIS_API_KEY"

4. Do NOT add a wire_api line. Current Codex removed wire_api = "chat" entirely and
   defaults to the Responses API, which is what MotAPI serves.
5. Ask me which shell profile to use, then export MOTAPIS_API_KEY=sk-mot-THAY_BANG_KHOA_CUA_BAN there.
6. Verify: run  codex doctor  and show me the output. It should say "config loaded" and
   print the model line; if the key is missing it states that explicitly.

CLAUDE MODELS WORK HERE TOO (since 26/09/2026): claude-sonnet-5, claude-opus-4-8,
claude-opus-5-5 and the other claude-* models run in Codex over the Responses API.
Other working models on this path include gpt-5.6-terra, gpt-6-astra, gpt-5.3-codex,
grok-4.6, glm-5.3, gemini-3.1-pro-preview, deepseek-v4-pro and qwen3.8-max.

Expected, not an error: Codex prints "Model metadata for ... not found" every session,
because it only knows OpenAI's own model catalogue. It still works.

IMPORTANT
- Never print, echo, or commit the API key.
- Do not change any other model provider already in config.toml.

You are configuring OpenCode CLI to use MotAPI (motapis.com).

Use these EXACT values:
  provider id = motapis
  baseURL     = https://motapis.com/v1
  model       = claude-opus-4-8

Steps:
1. Back up opencode.json if it already exists. Show me the diff before you write.
2. Create or merge opencode.json (project folder, or the global config) so it contains:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "motapis": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "MotAPI",
      "options": {
        "baseURL": "https://motapis.com/v1",
        "apiKey": "{env:MOTAPIS_API_KEY}"
      },
      "models": { "claude-opus-4-8": { "name": "Claude Opus 4.8" } }
    }
  }
}

   Keep every other provider already in the file unchanged.
3. Do NOT put the raw key in this file. Tell me to run /connect inside OpenCode and pick
   the "motapis" provider -- that is where the key is stored.
4. Verify: start opencode and check that motapis/claude-opus-4-8 appears in the model list.

IMPORTANT
- Never print, echo, or commit the API key.
- The provider id in opencode.json must match the one you select in /connect exactly.

You are configuring the Continue extension to use MotAPI (motapis.com).

Use these EXACT values:
  provider = openai            (OpenAI-compatible, not "anthropic")
  apiBase  = https://motapis.com/v1
  apiKey   = sk-mot-THAY_BANG_KHOA_CUA_BAN
  model    = claude-opus-4-8

Steps:
1. Find Continue's config file. It is usually ~/.continue/config.yaml -- if it is not
   there, check the path shown in Continue's own settings before creating a new one.
2. Back it up, then show me the diff before you write.
3. Append this model, keeping every model already listed:

models:
  - name: claude-opus-4-8
    provider: openai
    apiBase: https://motapis.com/v1
    apiKey: sk-mot-THAY_BANG_KHOA_CUA_BAN
    model: claude-opus-4-8

4. Verify: open the model picker in Continue -- the new model must appear in the list.

IMPORTANT
- Never print, echo, or commit the API key, and keep this file out of any git repository.
- Do not remove or edit any model or provider that is already configured.