Build and running steps

[lonnietc]

Hello,

I am researching MoE (Mixture-of-Experts) and MoA (Mixture-of-Agents) in Javascript and came across your site.

Can you please tell me the build and run steps so that I can run it locally?

Actually what I am seeking is how to actually run this in a browser from a web server but not from Node (NPM) if that is possible.

Thanks

[charleywolf]

1. Clone the repository
2. Add API keys to a .env file
3. npm install
4. npm run dev
5. Navigate to localhost:3000 or whatever port is listed in the terminal
   NPM is required to run this code

[lonnietc]

Thanks and I'm working on it now.

Actually there is no ".env" file or ".env_sample" in the drop as well.
What is the format, of course without your actual API keys?

Thanks again

[charleywolf]

I'm actually not sure. It has been quite a while. I assume it would be the standard ones: OPENAI_API_KEY, etc.

If you could check my code and figure this out and open a PR adding a .env.example and documentation on running it that would be much appreciated.

[lonnietc]

Hello,

I am working on the MoA today and am wondering if you remember it needing a POSTGRESQL database as I ran into an error when I went to the http://localhost:3000

Error: 
Invalid `prisma.chat.findMany()` invocation:


error: Environment variable not found: POSTGRES_PRISMA_URL.
  -->  schema.prisma:7
   | 
 6 |   provider  = "postgresql"
 7 |   url       = env("POSTGRES_PRISMA_URL")
   | 

Validation Error Count: 1

[charleywolf]

Oh yes, it does.

…

On Fri, Jan 16, 2026 at 7:17 AM LonnieTC ***@***.***> wrote: *lonnietc* left a comment (charleywolf/mixture#4) <#4 (comment)> Hello, I am working on the MoA today and am wondering if you remember it needing a POSTGRESQL database as I ran into an error when I went to the http://localhost:3000 Error: Invalid `prisma.chat.findMany()` invocation: error: Environment variable not found: POSTGRES_PRISMA_URL. --> schema.prisma:7 | 6 | provider = "postgresql" 7 | url = env("POSTGRES_PRISMA_URL") | Validation Error Count: 1 — Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJ2TXQYYOP7OKYCZ54LZOO34HDJEHAVCNFSM6AAAAACRZTYHZOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONJZG43TMMJRGA> . You are receiving this because you commented.Message ID: ***@***.***>

[lonnietc]

Hi Charley,

Actually, I just found it and am now working on having it run SQLITE locally for testing but am making some progress and hope to have it running soon.

If I understand the MoA design, then you have 4 LLMs with the "TOGETHER" LLM being the aggregator LLM. The other 3 are called with the use query and the results from them are fed into the TOGETHER LLM to give a result back to the user:

# Together AI is often the primary provider for MoA implementations
TOGETHER_API_KEY="your_together_ai_key_here"

# Optional: Add these if you plan to use OpenAI or Groq models
OPENAI_API_KEY="sk-..."
GROQ_API_KEY="gsk_..."
ANTHROPIC_API_KEY="sk-ant-..."

Does this sound correct to you?

I am working out the .env file and should have it soon.
Thanks again