Your key, inside your own tools

1. Open Help → Troubleshooting → Enable Developer Mode. After the app restarts, choose Developer → Configure Third-Party Inference…. On Windows the ☰ menu is in the top-left of the sign-in screen.
2. Set Inference provider to Gateway. Then fill in Gateway credentials as in the picture:

   

   ✓ https://api.mawen8899.com1 sk-ntm-…2 ✓

   1. Gateway base URL: https://api.mawen8899.com
   2. Gateway API key: paste your own key (it starts with sk-ntm-).
   3. Credential kind must be Static API key and Gateway auth scheme bearer. Leave the other fields empty.
3. Further down the same page, in Models, click Add model. Leave the other settings there as they are.

   

   Click →

4. Fill in the box that opens as in the picture:

   

   claude-opus-4-81 Opus 4.82 opus3 4

   1. Model ID: claude-opus-4-8 — exactly this; it is what reaches the server. Short names like opus-4.8 do not work.
   2. Display name: the name you see in the picker; left empty, the app makes one from the ID.
   3. Tier alias: the model's family, here opus.
   4. Add: for the next model.
   Leave Offer 1M-context variant off and Max effort empty.
5. Add as many models as you like; copy their IDs from the Model IDs list. Tier alias is the family named in the ID: claude-opus-… → opus, claude-sonnet-… → sonnet, claude-fable-… → fable. The first model in the list is the one the app starts with. Only Claude models belong in this screen: GPT and the other models speak the OpenAI format, so use one of the other tools for those.
6. Click Apply locally — the app saves the configuration and relaunches. Open the Code tab, pick one of the models you added and send a test message. All of them sit in the same picker, so you can switch between them whenever you like.

1. Open the settings file — create it if it is not there yet:
   macOS / Linux~/.claude/settings.json Windows%USERPROFILE%\.claude\settings.json
2. Put the address, your key and your models in its env block:

   {
     "env": {
       "ANTHROPIC_BASE_URL": "https://api.mawen8899.com",
       "ANTHROPIC_AUTH_TOKEN": "sk-ntm-…",
       "ANTHROPIC_MODEL": "claude-opus-4-8",
       "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4-8",
       "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5",
       "ANTHROPIC_DEFAULT_FABLE_MODEL": "claude-fable-5-1",
       "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-5-5"
     }
   }

   If the file already exists, add the env block to it instead of replacing the file.
3. What the model lines mean:

   ANTHROPIC_MODEL                  the model claude starts with
   ANTHROPIC_DEFAULT_OPUS_MODEL     the model behind the name "opus"
   ANTHROPIC_DEFAULT_SONNET_MODEL   the model behind "sonnet"
   ANTHROPIC_DEFAULT_FABLE_MODEL    the model behind "fable"
   ANTHROPIC_DEFAULT_HAIKU_MODEL    the model behind "haiku" — and the one
                                    used for small background requests

   Every value is a full model ID from the model table below. Keep "haiku" on Haiku 5.5, the cheapest model: titles and summaries run on it in the background, and without that line they go to your main — more expensive — model. To put a different model behind a name, change its line, for example "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5-5".
4. Run claude. Switch models whenever you like, without leaving the session:

   /model sonnet            → claude-sonnet-5
   /model fable             → claude-fable-5-1
   /model opus              → claude-opus-4-8
   /model haiku             → claude-haiku-5-5
   /model claude-opus-5-5   → any full model ID works as well

   Or start straight on one: claude --model sonnet. The file is read at startup, so restart claude after you edit it.

1. Open Settings (Cmd/Ctrl+Shift+J) → Models.
2. Expand OpenAI API Key and paste your key: sk-ntm-…
3. Enable Override OpenAI Base URL and enter https://api.mawen8899.com/v1
4. Click + Add model and enter a model ID, for example gpt-5.6-luna.
5. Click Verify — Cursor sends a test request and saves the key and the base URL.

1. Create the Codex configuration file:
   macOS / Linux~/.codex/config.toml Windows%USERPROFILE%\.codex\config.toml

   model = "gpt-5.6-luna"
   model_provider = "mawen"
   
   [model_providers.mawen]
   name = "Mawen API"
   base_url = "https://api.mawen8899.com/v1"
   env_key = "MAWEN_API_KEY"
   wire_api = "responses"

   To use another model, put any ID from the GPT group of the model table on the model line.
2. Put the key in the environment variable the file names.
   macOS / Linux — this terminal only

   export MAWEN_API_KEY="sk-ntm-…"

   Windows (PowerShell) — this terminal only

   $env:MAWEN_API_KEY = "sk-ntm-…"

3. Make it permanent, so a new terminal — and the Codex app — still has the key.
   Windows (PowerShell) — run once

   [System.Environment]::SetEnvironmentVariable('MAWEN_API_KEY', 'sk-ntm-…', 'User')

   Then close the terminal and open a new one.
   macOS — once, in Finder
   1. In Terminal, type touch ~/.zshrc and press Return. It creates the file if it is missing and changes nothing if it is already there.
   2. Open Finder and press ⌘ Cmd + Shift + G.
   3. Type ~/.zshrc in the box and press Return — the file comes up selected.
   4. Right-click it → Open With → TextEdit.
   5. Add this line at the very bottom, with your own key in place of sk-ntm-…:

      export MAWEN_API_KEY="sk-ntm-…"

   6. Save with ⌘ Cmd + S. Quit Terminal completely (⌘ Cmd + Q) and open it again.
   Linux — run once

   echo 'export MAWEN_API_KEY="sk-ntm-…"' >> ~/.bashrc

   Then close the terminal and open a new one.
4. Using the Codex desktop app? Quit it completely and open it again — it reads the key only when it starts. To quit it, right-click its icon next to the clock → Quit. To quit it, press ⌘ Cmd + Q while it is in front.

1. Create the configuration file and open it; run this in your terminal:
   macOS

   mkdir -p ~/.config/opencode && touch ~/.config/opencode/opencode.json && open -t ~/.config/opencode/opencode.json

   Linux

   mkdir -p ~/.config/opencode && nano ~/.config/opencode/opencode.json

   Windows (PowerShell)

   mkdir -Force "$HOME\.config\opencode" | Out-Null; notepad "$HOME\.config\opencode\opencode.json"

2. Put the provider in it, with your own key in place of sk-ntm-…, and save:

   {
     "$schema": "https://opencode.ai/config.json",
     "model": "mawen/gpt-6-luna",
     "provider": {
       "mawen": {
         "npm": "@ai-sdk/openai-compatible",
         "name": "Mawen API",
         "options": {
           "baseURL": "https://api.mawen8899.com/v1",
           "apiKey": "sk-ntm-…"
         },
         "models": {
           "gpt-6-luna": {
             "name": "GPT 6 Luna",
             "attachment": true,
             "modalities": { "input": ["text", "image"], "output": ["text"] },
             "limit": { "context": 200000, "output": 32000 }
           },
           "claude-opus-5-5": {
             "name": "Opus 5.5",
             "attachment": true,
             "modalities": { "input": ["text", "image"], "output": ["text"] },
             "limit": { "context": 200000, "output": 32000 }
           }
         }
       }
     }
   }

   This format works on both OpenCode 1.x and 2.x. For another model, add a block like these; the IDs are in the Model IDs list. If the file already has other settings, add these lines to them instead of replacing the file.
3. Restart OpenCode and open /models: your models are under Mawen API. If they are not there, OpenCode did not load the file: check the file's name and folder first, then look for a typing mistake. OpenCode skips a broken file without a word, so fix it before you go on. Keep /v1 in baseURL.

1. Open the Cline panel → ⚙.
2. API Provider → OpenAI Compatible.
3. Base URL → https://api.mawen8899.com/v1
4. API Key → sk-ntm-…
5. Model ID → type an ID from the model table. The list is not fetched automatically, so enter it by hand, then click Verify.

1. ⚙ → Providers tab → Custom provider at the bottom.
2. Set the Provider ID (mawen) and the Display name (Mawen API).
3. Provider API → OpenAI Compatible.
4. Enter the Base URL https://api.mawen8899.com/v1 and the API key sk-ntm-…
5. In Models, enter an ID from the model table. For a custom provider, set Context Window and Max Output Tokens in Model Configuration.

1. Export the key.
   macOS / Linux

   export MAWEN_API_KEY="sk-ntm-…"

   Windows (PowerShell)

   $env:MAWEN_API_KEY = "sk-ntm-…"

2. Add a custom model to settings.json:
   macOS / Linux~/.factory/settings.json Windows%USERPROFILE%\.factory\settings.json

   {
     "customModels": [
       {
         "model": "claude-opus-4-8",
         "displayName": "Mawen Opus 4.8",
         "baseUrl": "https://api.mawen8899.com/v1",
         "apiKey": "${MAWEN_API_KEY}",
         "provider": "generic-chat-completion-api",
         "maxOutputTokens": 16384
       }
     ]
   }

3. Run droid, execute /model and pick the entry from Custom models. settings.json uses camelCase fields (do not mix it with the legacy config.json / snake_case), the provider is generic-chat-completion-api, and ${VAR} only works in settings.json.

1. Create the models file and open it; run this in your terminal:
   macOS

   mkdir -p ~/.pi/agent && touch ~/.pi/agent/models.json && open -t ~/.pi/agent/models.json

   Linux

   mkdir -p ~/.pi/agent && nano ~/.pi/agent/models.json

   Windows (PowerShell)

   mkdir -Force "$HOME\.pi\agent" | Out-Null; notepad "$HOME\.pi\agent\models.json"

2. Put the provider in it, with your own key in place of sk-ntm-…, and save:

   {
     "providers": {
       "mawen": {
         "name": "Mawen API",
         "baseUrl": "https://api.mawen8899.com/v1",
         "api": "openai-completions",
         "apiKey": "sk-ntm-…",
         "models": [
           {
             "id": "gpt-6-luna",
             "name": "GPT 6 Luna",
             "input": ["text", "image"],
             "contextWindow": 200000,
             "maxTokens": 32000
           },
           {
             "id": "claude-opus-5-5",
             "name": "Opus 5.5",
             "input": ["text", "image"],
             "contextWindow": 200000,
             "maxTokens": 32000
           }
         ]
       }
     }
   }

   For another model, add a block like these; the IDs are in the Model IDs list. If the file already has other providers, add only the mawen block next to them.
3. In your project folder, run pi, type /model, move to a model marked [mawen] and press Ctrl+S: it is selected and stays your default.

1. Create the models file and open it; run this in your terminal:
   macOS

   mkdir -p ~/.omp/agent && touch ~/.omp/agent/models.yml && open -t ~/.omp/agent/models.yml

   Linux

   mkdir -p ~/.omp/agent && nano ~/.omp/agent/models.yml

   Windows (PowerShell)

   mkdir -Force "$HOME\.omp\agent" | Out-Null; notepad "$HOME\.omp\agent\models.yml"

2. Put the provider in it, with your own key in place of sk-ntm-…, and save:

   providers:
     mawen:
       baseUrl: https://api.mawen8899.com/v1
       api: openai-completions
       apiKey: sk-ntm-…
       models:
         - id: gpt-6-luna
           name: Mawen GPT 6 Luna
           reasoning: false
           input: [text, image]
           contextWindow: 200000
           maxTokens: 32000
         - id: claude-opus-5-5
           name: Mawen Opus 5.5
           reasoning: false
           input: [text, image]
           contextWindow: 200000
           maxTokens: 32000

   For another model, add a block like these; the IDs are in the Model IDs list. If the file already has a providers: line, add only the mawen block under it.
3. Check it: omp models mawen must list your models.
4. In your project folder, run omp. On the first launch, press Esc at Set up your providers; at Choose your default model, type mawen and pick a model. To change it later, type /model, find the model and assign it to Default.

1. Open the settings file with hermes config edit. It lives here:
   macOS / Linux~/.hermes/config.yaml Windows%LOCALAPPDATA%\hermes\config.yaml
2. Find the model: line at the far left of the file. Delete it and all the indented lines under it, paste this in its place with your own key instead of sk-ntm-…, and save:

   model:
     provider: custom:mawen
     default: gpt-6-luna
   
   providers:
     mawen:
       name: Mawen API
       api: https://api.mawen8899.com/v1
       api_key: sk-ntm-…
       transport: chat_completions
       default_model: gpt-6-luna
       discover_models: false
       models:
         gpt-6-luna:
           context_length: 200000
           supports_vision: true
         claude-opus-5-5:
           context_length: 200000
           supports_vision: true

   Only one model: and one providers: line may start at the far left of the file; lines with spaces or # in front do not count. With a second one, Hermes cannot read the file and your settings do not apply. For another model, add it under models: the same way; the IDs are in the Model IDs list.
3. Test it: hermes chat --oneshot -q "Reply with exactly OK". A reply means it works.
4. In your project folder, run hermes. /model opens the list: Mawen API, then the model. That choice lasts for the session; to keep it, type /model claude-opus-5-5 --global.

1. If you have not set OpenClaw up yet, run openclaw setup --baseline first.
2. Create a file for the settings and open it; run this in your terminal:
   macOS

   touch mawen.patch.json && open -t mawen.patch.json

   Linux

   nano mawen.patch.json

   Windows (PowerShell)

   notepad mawen.patch.json

3. Put this in it, with your own key in place of sk-ntm-…, and save:

   {
     "models": {
       "mode": "merge",
       "providers": {
         "mawen": {
           "baseUrl": "https://api.mawen8899.com/v1",
           "apiKey": "sk-ntm-…",
           "api": "openai-completions",
           "models": [
             {
               "id": "gpt-6-luna",
               "name": "Mawen GPT 6 Luna",
               "input": ["text", "image"],
               "contextWindow": 200000,
               "maxTokens": 32000
             },
             {
               "id": "claude-opus-5-5",
               "name": "Mawen Opus 5.5",
               "input": ["text", "image"],
               "contextWindow": 200000,
               "maxTokens": 32000
             }
           ]
         }
       }
     },
     "agents": {
       "defaults": {
         "model": { "primary": "mawen/gpt-6-luna" },
         "models": {
           "mawen/gpt-6-luna": { "alias": "mw-luna" },
           "mawen/claude-opus-5-5": { "alias": "mw-opus" }
         }
       }
     }
   }

   For another model, add a block to the models list and a line with its alias under agents; the IDs are in the Model IDs list.
4. Add it to your settings: openclaw config patch --file ./mawen.patch.json. This adds these lines and leaves your other settings as they are. Then delete mawen.patch.json: your key is in it.
5. Run openclaw chat. Switch models with /model mw-opus or /model mw-luna. To change the default: openclaw models set mawen/claude-opus-5-5.

1. Run kimi --version: it must start with 2. Version 1.x is the old tool and reads a different file.
2. Create the settings file and open it; run this in your terminal:
   macOS

   mkdir -p ~/.kimi-code && touch ~/.kimi-code/config.toml && open -t ~/.kimi-code/config.toml

   Linux

   mkdir -p ~/.kimi-code && nano ~/.kimi-code/config.toml

   Windows (PowerShell)

   mkdir -Force "$HOME\.kimi-code" | Out-Null; notepad "$HOME\.kimi-code\config.toml"

3. Put this in it, with your own key in place of sk-ntm-…, and save:

   default_model = "mawen/gpt-6-luna"
   
   [providers.mawen]
   type = "openai"
   base_url = "https://api.mawen8899.com/v1"
   api_key = "sk-ntm-…"
   
   [models."mawen/gpt-6-luna"]
   provider = "mawen"
   model = "gpt-6-luna"
   display_name = "Mawen GPT 6 Luna"
   max_context_size = 200000
   capabilities = ["tool_use", "image_in"]
   
   [models."mawen/claude-opus-5-5"]
   provider = "mawen"
   model = "claude-opus-5-5"
   display_name = "Mawen Opus 5.5"
   max_context_size = 200000
   capabilities = ["tool_use", "image_in"]

   The default_model line must stay at the very top, above the first line that starts with [, and only once. For another model, add a block like these; the IDs are in the Model IDs list.
4. Run kimi in your project folder. You do not need /login: that is for Kimi accounts only. Switch models with /model. On Windows, Git for Windows must be installed first: Kimi Code does not run without it.

1. Start it with npx @deepseek-ai/dsh webnpx.cmd @deepseek-ai/dsh web (it needs Node.js 24.9 or newer); it opens in your browser. Click Continue on Preview Notice, then Configure later on Add an API key to get started. In the desktop app, go straight to step 2.
2. Open Settings → Models → Add model provider and switch to Custom model API.
3. Provider ID → mawen, Display name → Mawen API.
4. Base URL → https://api.mawen8899.com/v1, API protocol → OpenAI Chat Completions.
5. API key → sk-ntm-…
6. Click Fetch available models, tick the models you want and click Add selected. Then click Create provider.
7. Pick your project with Choose workspace, then choose the model under Mawen API in the message box.

Any application that speaks the OpenAI or the Anthropic API works: give it the address, the key and a model ID.

OpenAI format:     https://api.mawen8899.com/v1
Anthropic format:  https://api.mawen8899.com   (without /v1)
Authorization:     Bearer sk-ntm-…

# OpenAI SDK (Python)
from openai import OpenAI
client = OpenAI(base_url="https://api.mawen8899.com/v1", api_key="sk-ntm-…")

# curl
curl https://api.mawen8899.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ntm-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"Hello"}],"stream":true}'

Both formats accept x-api-key and Authorization: Bearer, and both draw from the same balance.