Yesterday was a milestone for our OpenClaw stack: we wired Vertex AI directly into openclaw.json and made Gemini models first class citizens across multiple regions. The goal was simple but ambitious: get reliable, OpenAI compatible routing for Gemini 2.5 while keeping fallbacks deterministic and performance consistent.
Why This Integration Mattered
We wanted a setup that could survive network hiccups and regional outages without manual intervention. By routing through Google’s OpenAI compatible endpoints in Vertex AI, we get a consistent request and response shape while still benefiting from Gemini’s strength in speed and multimodality.
- Reliability: Multi region endpoints reduce single region failure risk.
- Consistency: OpenAI compatible APIs make model swapping easy.
- Control: Explicit fallback chains keep workloads stable.
What We Implemented
We added several Vertex AI providers (global and regional) and registered Gemini 2.5 Flash/Pro. We also updated our model priority chain so Gemini 2.5 Flash stays the primary choice, with carefully ordered fallbacks.
Key Configuration (Sanitized Excerpt)
{
"models": {
"providers": {
"vertex-global": {
"baseUrl": "https://aiplatform.googleapis.com/v1beta1/projects//locations/global/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Global)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-global-pro": {
"baseUrl": "https://aiplatform.googleapis.com/v1beta1/projects//locations/global/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (Global)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-us1": {
"baseUrl": "https://us-central1-aiplatform.googleapis.com/v1beta1/projects//locations/us-central1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (US Iowa)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-us2": {
"baseUrl": "https://us-east4-aiplatform.googleapis.com/v1beta1/projects//locations/us-east4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (US Virginia)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-eu1": {
"baseUrl": "https://europe-west1-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU Belgium)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-eu2": {
"baseUrl": "https://europe-west4-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU Netherlands)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-eu3": {
"baseUrl": "https://europe-west2-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west2/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU London)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-asia1": {
"baseUrl": "https://asia-southeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-southeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Asia Singapore)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-asia2": {
"baseUrl": "https://asia-northeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-northeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Asia Tokyo)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-india": {
"baseUrl": "https://asia-south1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-south1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (India Mumbai)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-global": {
"baseUrl": "https://aiplatform.googleapis.com/v1beta1/projects//locations/global/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Global)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-us1": {
"baseUrl": "https://us-central1-aiplatform.googleapis.com/v1beta1/projects//locations/us-central1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (US Iowa)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-us2": {
"baseUrl": "https://us-east4-aiplatform.googleapis.com/v1beta1/projects//locations/us-east4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (US Virginia)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-eu1": {
"baseUrl": "https://europe-west1-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU Belgium)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-eu2": {
"baseUrl": "https://europe-west4-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU Netherlands)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-eu3": {
"baseUrl": "https://europe-west2-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west2/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (EU London)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-asia1": {
"baseUrl": "https://asia-southeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-southeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Asia Singapore)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-asia2": {
"baseUrl": "https://asia-northeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-northeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (Asia Tokyo)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3-india": {
"baseUrl": "https://asia-south1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-south1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-flash",
"name": "Gemini 2.5 Flash (India Mumbai)",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-global": {
"baseUrl": "https://aiplatform.googleapis.com/v1beta1/projects//locations/global/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (Global)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-us1": {
"baseUrl": "https://us-central1-aiplatform.googleapis.com/v1beta1/projects//locations/us-central1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (US Iowa)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-us2": {
"baseUrl": "https://us-east4-aiplatform.googleapis.com/v1beta1/projects//locations/us-east4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (US Virginia)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-eu1": {
"baseUrl": "https://europe-west1-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (EU Belgium)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-eu2": {
"baseUrl": "https://europe-west4-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west4/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (EU Netherlands)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-eu3": {
"baseUrl": "https://europe-west2-aiplatform.googleapis.com/v1beta1/projects//locations/europe-west2/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (EU London)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-asia1": {
"baseUrl": "https://asia-southeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-southeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (Asia Singapore)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-asia2": {
"baseUrl": "https://asia-northeast1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-northeast1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (Asia Tokyo)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
},
"vertex-g3pro-india": {
"baseUrl": "https://asia-south1-aiplatform.googleapis.com/v1beta1/projects//locations/asia-south1/endpoints/openapi",
"apiKey": "",
"api": "openai-completions",
"models": [
{
"id": "google/gemini-2.5-pro",
"name": "Gemini 2.5 Pro (India Mumbai)",
"reasoning": true,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 1000000,
"maxTokens": 65536
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "vertex-g3-global/google/gemini-2.5-flash",
"fallbacks": [
"vertex-g3-global/google/gemini-2.5-flash",
"vertex-g3-us1/google/gemini-2.5-flash",
"vertex-g3-us2/google/gemini-2.5-flash",
"vertex-g3-eu1/google/gemini-2.5-flash",
"vertex-g3-eu2/google/gemini-2.5-flash",
"vertex-g3-eu3/google/gemini-2.5-flash",
"vertex-g3-asia1/google/gemini-2.5-flash",
"vertex-g3-asia2/google/gemini-2.5-flash",
"vertex-g3-india/google/gemini-2.5-flash",
"vertex-g3pro-global/google/gemini-2.5-pro",
"vertex-g3pro-us1/google/gemini-2.5-pro",
"vertex-g3pro-us2/google/gemini-2.5-pro",
"vertex-g3pro-eu1/google/gemini-2.5-pro",
"vertex-g3pro-eu2/google/gemini-2.5-pro",
"vertex-g3pro-eu3/google/gemini-2.5-pro",
"vertex-g3pro-asia1/google/gemini-2.5-pro",
"vertex-g3pro-asia2/google/gemini-2.5-pro",
"vertex-g3pro-india/google/gemini-2.5-pro"
]
}
}
}
} 
What We Learned
- Model ordering matters: It defines both reliability and cost behavior.
- OpenAI compatible endpoints are a big win: We can swap providers without rewriting our clients.
- Fallbacks must be explicit: This keeps failures from degrading overall quality.
Next Steps
We plan to add an automated health check across regions, and keep a close eye on log routing so error heavy traces do not degrade model context.
In short: we now have a robust, multi region Vertex AI layer inside OpenClaw, and it is already paying off.
Discover more from TheFlipbit
Subscribe to get the latest posts to your email.