Setting Up MiniMax-M3 in Prime Agent

If you're trying to use the MiniMax M3 model with Prime Agent and getting cryptic errors like `No API provider registered for api: anthropic` or `Model not found: minimax/MiniMax-M3`, this guide walks you through the fix.


## The Problem


Prime Agent's `models.json` configuration file has a few gotchas that aren't obvious from the documentation. After a few rounds of debugging, here are the three things that need to be correct:


1. The **`api` field** must be one of the values Prime Agent actually recognizes (`anthropic-messages`, `openai-completions`, `openai-responses`).

2. The **`baseUrl`** must match the API protocol you're using.

3. The **provider name** in `models.json` must match the one referenced in `settings.json`.


Get any of these wrong and you get a frustrating error that doesn't immediately point to the cause.


## Understanding MiniMax's API


[MiniMax](https://platform.minimax.io) exposes two API protocols, each on a different base URL:


| Protocol      | Base URL                          | Prime Agent `api` value |

|---------------|-----------------------------------|-------------------------|

| Anthropic API | `https://api.minimax.io/anthropic` | `anthropic-messages`    |

| OpenAI API    | `https://api.minimax.io/v1`        | `openai-completions` or `openai-responses` |


Pick whichever protocol your tooling supports. For Prime Agent's Anthropic-compatible model, use the Anthropic endpoint.


## The Fix


Edit `~/.prime/agent/models.json`:


```json

{

  "providers": {

    "minimax": {

      "name": "MiniMax",

      "baseUrl": "https://api.minimax.io/anthropic",

      "apiKey": "your-subscription-key-here",

      "api": "anthropic-messages",

      "authHeader": true,

      "models": [

        {

          "id": "MiniMax-M3",

          "name": "MiniMax-M3",

          "reasoning": true,

          "input": ["text", "image"],

          "contextWindow": 1000000,

          "maxTokens": 32000,

          "compat": {}

        }

      ]

    }

  }

}

```


Key points:


- `api` is `"anthropic-messages"`, not `"anthropic"`. Prime Agent only recognizes the full name.

- `baseUrl` includes the `/anthropic` path. Without it, requests hit the wrong endpoint and return 404.

- The provider key (`minimax`) is what you'll reference in `settings.json` and when switching models.


Then update `~/.prime/agent/settings.json` to use it:


```json

{

  "defaultProvider": "minimax",

  "defaultModel": "MiniMax-M3"

}

```


Run `/reload` in Prime Agent and you should be good to go.


## Step-by-Step


1. Get a subscription key from the [MiniMax Token Plan](https://platform.minimax.io) page.

2. Create or edit `~/.prime/agent/models.json` with the configuration above.

3. Make sure `settings.json` references the same provider name (`minimax`).

4. Run `/reload` in Prime Agent.

5. Switch to the model with `minimax/MiniMax-M3`.


## Verifying the API Works


Before assuming the issue is on Prime Agent's side, test MiniMax directly:


```bash

curl -X POST https://api.minimax.io/anthropic/v1/messages \

  -H "x-api-key: $MINIMAX_API_KEY" \

  -H "anthropic-version: 2023-06-01" \

  -H "content-type: application/json" \

  -d '{

    "model": "MiniMax-M3",

    "max_tokens": 100,

    "messages": [{"role": "user", "content": "Hello"}]

  }'

```


A successful response looks like:


```json

{

  "id": "...",

  "type": "message",

  "role": "assistant",

  "model": "MiniMax-M3",

  "content": [{"text": "Hello!", "type": "text"}],

  ...

}

```


If this returns 200, MiniMax is reachable and your API key is valid. Any Prime Agent errors after that are configuration issues.


## Troubleshooting


| Error | Cause | Fix |

|-------|-------|-----|

| `No API provider registered for api: anthropic` | `api` field is set to `"anthropic"` | Change to `"anthropic-messages"` |

| `Model not found: minimax/MiniMax-M3` | Provider name in `models.json` doesn't match `settings.json` | Make sure both use `minimax` (or whatever name you pick) |

| 404 from `https://api.minimax.io/v1/messages` | `baseUrl` is missing the `/anthropic` path | Use `https://api.minimax.io/anthropic` for the Anthropic protocol |

| `401 Unauthorized` | Wrong or expired API key | Get a new subscription key from the Token Plan page |


## Why These Specific Values?


Prime Agent's [extensions documentation](https://github.com/PrimeIntellect-ai/prime-agent/blob/main/packages/coding-agent/docs/extensions.md) shows three valid `api` values for `registerProvider()`: `openai-completions`, `openai-responses`, and `anthropic-messages`. The shorter `anthropic` value is what the official Anthropic SDK uses, but Prime Agent requires the full name.


The `baseUrl` quirk is documented in MiniMax's [Other Tools guide](https://platform.minimax.io/docs/token-plan/other-tools): the Anthropic-compatible protocol lives at `/anthropic`, while OpenAI-compatible lives at `/v1`. Using the bare hostname (`https://api.minimax.io`) doesn't route to either.


## References


- [Prime Agent — Extensions documentation](https://github.com/PrimeIntellect-ai/prime-agent/blob/main/packages/coding-agent/docs/extensions.md)

- [MiniMax — Anthropic API reference](https://platform.minimax.io/docs/api-reference/text-anthropic-api)

- [MiniMax — OpenAI API reference](https://platform.minimax.io/docs/api-reference/text-openai-api)

- [MiniMax — Other Tools configuration](https://platform.minimax.io/docs/token-plan/other-tools)

- [MiniMax — Token Plan](https://platform.minimax.io)


Comments

Post a Comment

Popular posts from this blog

Switching from Docker Desktop to Rancher Desktop on WSL2: Solving the Missing docker-credential-desktop.exe Error

Switching from Docker Desktop to Rancher Desktop on WSL2: Solving the Missing docker-credential-desktop.exe Error I’ve been happily running Kamal inside WSL2 on my home Windows desktop for a while. My original setup included an older Docker Desktop installation. Recently, I wanted to replicate this environment on a different machine, but I discovered that Docker Desktop had changed their licensing terms. Technically, I’m still covered by the new licensing terms, but I decided to try Rancher Desktop instead. Installing Rancher Desktop on WSL2 Installation was straightforward. I installed Rancher Desktop, configured it to integrate with WSL2, and disabled Kubernetes (since I don’t currently need it). My plan was to continue using the Docker CLI inside WSL2 alongside Rancher Desktop for container management. The Authentication Error However, when I was in WSL and tried to authenticate (for example, with Docker Hub via docker login), I ran into this error: ``` failed to store...
Read more

Running LXD/LXC on WSL2 with Ubuntu 20.04

Download Ubuntu 20.04 from the Microsoft Store sudo apt-get update sudo apt-get upgrade cd ~ git clone https://github.com/DamionGans/ubuntu-wsl2-systemd-script.git cd ubuntu-wsl2-systemd-script/ bash ubuntu-wsl2-systemd-script.sh sudo apt-get install snap (exit the shell and reopen it) Source: https://github.com/damionGans/ubuntu-wsl2-systemd-script To the IPv6 question I answer "none". $ sudo lxd init 2020/08/01 15:55:13 usbid: failed to load: open /usr/share/misc/usb.ids: no such file or directory Would you like to use LXD clustering? (yes/no) [default=no]: Do you want to configure a new storage pool? (yes/no) [default=yes]: Name of the new storage pool [default=default]: Name of the storage backend to use (ceph, btrfs, dir, lvm) [default=btrfs]: Create a new BTRFS pool? (yes/no) [default=yes]: Would you like to use an existing block device? (yes/no) [default=no]: Size in GB of the new loop device (1GB minimum) [default=50GB]: Would you like to connect to a MAAS server? (...
Read more

Proxmox Remote administration via VNC

Image
A few times I've needed to communicate via VNC, for instance when I was changing network configuration and couldn't use SSH or when I tried out Windows. Communicating with VNC to Proxmox wasn't so easy for me as it seemed in the documentation. Basically I never really got the " /etc/inetd.conf " thing to work, but this made it work from my local machine. This blog post is an alternative to the openbsd-inetd solution In the following: 2.2.2.2 is the IP of my Proxmox host 100 is the VM-ID you picked when you created the virtual machine ======================================================================== ssh -f -L 25901:127.0.0.1:5910 root@ 2.2.2.2 'nc -l -p 5910 -c "qm vncproxy 100 "'; "/Applications/TigerVNC Viewer 1.2.0.app/Contents/MacOS/TigerVNC Viewer" localhost:25901 ======================================================================== Please note that I only got TigerVNC to work, since all the other VNCs ha...
Read more