Use MyTabulon with OpenClaw
Power OpenClaw agents with MyTabulon's OpenAI-compatible Coding Plan gateway. Register a custom provider, keep the API key in an environment variable, and preserve each model's native context.
Add the provider
Add MyTabulon under models.providers in your OpenClaw configuration. This example uses OpenAI Chat Completions, which is the broadest compatibility path for the multi-provider Coding Plan. It deliberately omits contextTokens, so OpenClaw does not impose a smaller effective context cap.
{
"env": {
"MYTABULON_API_KEY": "mtb_live_..."
},
"agents": {
"defaults": {
"model": { "primary": "mytabulon/maximo-atlas-preview" },
"models": {
"mytabulon/maximo-atlas-preview": { "alias": "Atlas" }
}
}
},
"models": {
"mode": "merge",
"providers": {
"mytabulon": {
"baseUrl": "https://api.mytabulon.com/v1",
"apiKey": "${MYTABULON_API_KEY}",
"api": "openai-completions",
"models": [{
"id": "maximo-atlas-preview",
"name": "Maximo Atlas Preview",
"reasoning": true,
"input": ["text", "image"],
"contextWindow": 262000,
"maxTokens": 65536
}]
}
}
}
}Keep secrets outside the config
For production, prefer OpenClaw's SecretRef or environment substitution instead of storing a live key in the JSON file. The inline env block above is easy for a first local test; move MYTABULON_API_KEY to ~/.openclaw/.env or your deployment secret manager before sharing or backing up the configuration.
Add more tier models
OpenClaw custom providers require a model entry for each model you want to select. Copy an entry, change id, name, contextWindow, input types, and reasoning based on GET /models, then add provider/model under agents.defaults.models. Do not set contextTokens unless you intentionally want a lower runtime cap.
curl https://api.mytabulon.com/v1/models \
-H "Authorization: Bearer $MYTABULON_API_KEY"
openclaw models
openclaw doctorUse Responses when you need it
The Coding Plan also supports POST /responses. Advanced OpenClaw setups can set api to openai-responses because the MyTabulon base URL implements /v1/responses. The openai-completions adapter remains the recommended default for the widest model compatibility.