How Configure Workers AI as models?

[kiran-4444]

I tried using this config:

{
    "meta": {
        "lastTouchedVersion": "2026.1.24-3",
        "lastTouchedAt": "2026-02-02T18:31:09.575Z"
    },
    "agents": {
        "defaults": {
            "model": {
                "primary": "workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast"
            }
        }
    },
    "models": {
        "mode": "merge",
        "providers": {
            "workers-ai": {
                "baseUrl": "https://api.cloudflare.com/client/v4/accounts/account_id/ai/v1",
                "apiKey": "key",
                "api": "openai-completions",
                "models": [
                    {
                        "id": "@cf/meta/llama-3.3-70b-instruct-fp8-fast",
                        "name": "Llama 3.3 70B"
                    },
                    {
                        "id": "@cf/meta/llama-3.1-8b-instruct",
                        "name": "Llama 3.1 8B"
                    },
                    {
                        "id": "@cf/mistral/mistral-7b-instruct-v0.2",
                        "name": "Mistral 7B"
                    },
                    {
                        "id": "@cf/qwen/qwen1.5-14b-chat-awq",
                        "name": "Qwen 1.5 14B"
                    },
                    {
                        "id": "@cf/deepseek-ai/deepseek-r1-distill-qwen-32b",
                        "name": "DeepSeek R1 32B"
                    }
                ]
            }
        }
    },
    "messages": {
        "ackReactionScope": "group-mentions"
    },
    "commands": {
        "native": "auto",
        "nativeSkills": "auto"
    },
    "channels": {
        "discord": {
            "token": "token",
            "allowBots": true,
            "groupPolicy": "allowlist",
            "actions": {
                "reactions": true,
                "emojiUploads": true,
                "messages": true,
                "memberInfo": true,
                "channelInfo": true
            },
            "guilds": {
                "guild": {}
            },
            "heartbeat": {
                "showOk": false,
                "showAlerts": true
            }
        }
    },
    "gateway": {
        "port": 18789,
        "mode": "local",
        "auth": {
            "token": "token"
        },
        "trustedProxies": [
            "10.1.0.0"
        ]
    },
    "plugins": {
        "entries": {
            "discord": {
                "enabled": true
            }
        }
    }
}

But somehow, upon saving this in Clawbot Control, the default model is changing back to opus 4.5. Am I doing something wrong here? sorry if this is naive, but I'm a bit new to cloudflare environment. Thanks!

[kwilson62]

You can change the model in the start-moltbot.sh starting on line 136, it directly writes the config file.

The issue here is you are trying to change the config inside the control UI, but when it restarts it wipes your changes with the config created by the startup script.

[booblean]

trying to do this now and not gateway to a third party, I want the bot to use a cloudflare derived model, but unsure how this is done.

[zentico]

You can change the model in the start-moltbot.sh starting on line 136, it directly writes the config file.

The issue here is you are trying to change the config inside the control UI, but when it restarts it wipes your changes with the config created by the startup script.

I'm struggling with the clawbot.json persistence and getting the AI Gateway to work to connect to Grok instead of Anthropic. References to Anthropic always creep back in and no requests are getting thru to the AI Gateway per its logs (despite the proper secrets setting it up). Any advice on this welcome (I should start a new Issue).

However I wanted to point out that if you have a R2 working properly, the modified config file should persist, so no changes needed to the start-moltbot.sh. These changes can be made by uploading a revised config file to the R2, or directly from the config in the Control UI. Am I understanding correctly?

[cri-st]

If there’s one thing they MUST improve here, it’s definitely the documentation for all of us who don’t want to use Anthropic’s Opus; we just want to use their AI Gateway, nothing unusual.

[DobreAi-Vicky]

Since there is no fix for so many days - am I being inappropriate claiming it looks like a nice deal between Anthropic and Cloudflare? ;)

[DobreBiuro]

You can change the model in the start-moltbot.sh starting on line 136, it directly writes the config file.

The issue here is you are trying to change the config inside the control UI, but when it restarts it wipes your changes with the config created by the startup script.

Did you succeed with overwriting the overwriter?

[booblean]

I have had no success with runing the bot with the above variables to ai gateway and not a third party provider, will there an update on how to deploy with AI gateway routing to shared billing through models offered by cloudflare?

[HexonBot]

this moltworker absolute fail, no way to change ai model like to qwen portal or any other what originaly openclaw provides