A 12B model. Our data. Our iron.

Gemma 4 12B IT QAT on GPUs we actually run. Point the OpenAI SDK at us, then give it 45,504 documents through the data layer. Small weights. Large retrieval.

Get an API key Read the docs
Gemma 4 12B
API id vllm-primary · 16k
14-node cluster
5 of them GPUs
45,504 documents
vector + keyword + graph
POST /v1/chat/completions 
curl https://api.solidrust.ai/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vllm-primary",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

vllm-primary is the alias. Weights: google/gemma-4-12B-it-qat-w4a16-ct

Used in production by

MyAshes.ai AI companion platform Aidiant.com AI council platform AmiNex.ai Game intelligence + companions SolidRusT.net The forge

What we already run on it

Our own products hit the same API you get a key for.

MyAshes.ai

AI companion platform

Companion chat on Gemma 4 12B, memory through the data layer instead of a context dump.

Aidiant.com

AI council platform

A council of specialists, one inference backend, agent tools when a member needs retrieval.

AmiNex.ai

Game intelligence + companions

EVE Online incursions, fittings, market — live game data next to the same chat and RAG stack.

SolidRusT.net

The forge

Parent shop. Not a Zone companion. This is where the other apps get built.

A 12B with a library

Gemma 4 12B on five GPUs. The data layer is 45k embedded docs, keyword search, and a knowledge graph on the same key. That is how a small model stops sounding small.

Gemma 4 12B

Chat on GPUs we run. API id vllm-primary, 16k context. OpenAI-compatible.

/v1/chat/completions

Qwen3 Embeddings

Qwen3-Embedding-0.6B. 1024-dim vectors, 32k input. Same key as chat.

/v1/embeddings

Data layer

45,504 documents in Milvus. 45,318 in MeiliSearch. Hybrid = vector + keyword + graph.

/data/v1/query/*

Agents

Six tools on a 12B: semantic, keyword/graph hybrid, memory. The model stays small; the retrieval does not.

/v1/agent/chat

OpenAPI

Interactive data-layer spec. Auth with the same key.

/data/docs

Console

Mint a key. Watch usage. Upgrade tiers.

console.solidrust.ai
GET /data/v1 — live document counts (2026-09-04)
Milvus (semantic) 45,504
MeiliSearch (keyword) 45,318
arXiv 42,337
Knowledge base 2,223
Wikipedia 361
EVE wiki + SDE 228
Internal docs (Pantheon, strategy, API) 326

Call it like OpenAI

Point the official SDK at our base URL. No new client required.

APIInteractive Playground
Try it live!

Get your API key from console.solidrust.ai

View cURL equivalent
curl -X POST https://api.solidrust.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "vllm-primary",
    "messages": [{"role": "user", "content": "Say hello and introduce yourself briefly."}],
    "max_tokens": 512,
    "temperature": 0.7
  }'
Python 
from openai import OpenAI

client = OpenAI(
    base_url="https://api.solidrust.ai/v1",
    api_key="your-api-key"
)

response = client.chat.completions.create(
    model="vllm-primary",
    messages=[
        {"role": "user", "content": "Hello!"}
    ],
    stream=True
)

for chunk in response:
    print(chunk.choices[0].delta.content, end="")
JavaScript 
import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.solidrust.ai/v1',
  apiKey: 'your-api-key',
});

const stream = await client.chat.completions.create({
  model: 'vllm-primary',
  messages: [
    { role: 'user', content: 'Hello!' }
  ],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content || '');
}

Endpoints

Endpoint Description
/v1/chat/completions Chat completions (OpenAI compatible)
/v1/embeddings Text embeddings (1024-dim)
/data/v1/query/semantic Semantic vector search
/data/v1/query/hybrid Hybrid search (vector + keyword + graph)
/v1/agent/chat Tool-enabled AI agent
/v1/models List available models
Full documentation →

Pricing

Keys and billing live in the console. These are the tiers the console sells — not a dedicated Kubernetes namespace.

Free

Shared inference. Get a key and call the API.

$0 /month
  • Chat completions (`vllm-primary`)
  • 10,000 requests / month
  • 1 API key
  • 100 requests / minute
Get a key

Pro

Embeddings plus more keys and headroom.

$29 /month
  • Everything in Free
  • Embeddings API
  • 5 API keys
  • Higher rate limits
Upgrade in console

Enterprise

Agents and the highest limits on this cluster.

$99 /month
  • Everything in Pro
  • Agent endpoints
  • 20 API keys
  • Unlimited monthly quota
Upgrade in console

Free-tier quotas are enforced by the console (10k requests / month, inference scope). Paid prices and remaining limits are whatever the console currently shows at checkout — do not treat this page as a Stripe invoice.