AI Services And Agent Stacks

Moltern's catalog includes AI-focused services that can be combined into chat systems, agent builders, model gateways, retrieval workflows, and evaluation stacks.

Common AI Building Blocks​

Building Block	Examples
Chat UI	Open WebUI, LibreChat, LobeChat, AnythingLLM
Agent or workflow builder	Flowise, Langflow
Model runtime or gateway	Ollama, LiteLLM, NewAPI
Vector database	Qdrant, Chroma, Weaviate
Observability and evaluation	Langfuse, Argilla
Search and translation	SearXNG, LibreTranslate
Labeling and data prep	Label Studio, Argilla

The exact catalog can change. Use the in-app service docs and service catalog as the source of truth.

Pattern 1: Private Chat UI​

Use this when a team needs a private AI chat interface.

Typical services:

• Chat UI service.
• Model runtime or external model provider.
• Optional model gateway.
• Optional vector database for knowledge retrieval.

Deployment flow:

1. Create an ai or production environment.
2. Deploy the model runtime or configure the external provider.
3. Deploy the chat UI.
4. Add required variables in the chat UI.
5. Open the generated URL.
6. Configure users and authentication inside the chat UI.
7. Add a custom domain only after access control is ready.

Pattern 2: Agent Builder​

Use this when teams want visual workflows or agent chains.

Typical services:

• Flowise or Langflow.
• Model gateway or model runtime.
• Vector database.
• Optional observability service.

Deployment flow:

1. Deploy the vector database if retrieval is needed.
2. Deploy the model gateway or runtime.
3. Deploy the agent builder.
4. Add provider credentials as variables.
5. Create a test workflow.
6. Confirm logs and service status.
7. Document which environment owns the stack.

Pattern 3: Model Gateway​

Use a model gateway when multiple apps or tools should call AI providers through one managed endpoint.

Benefits:

• Centralized provider configuration.
• Easier spend tracking inside the gateway.
• Common authentication layer.
• One endpoint for several AI tools.

Do not expose a model gateway publicly without authentication and rate limits inside the gateway.

Pattern 4: Retrieval Stack​

Use this when users need chat or agents over documents.

Typical components:

• Chat or agent UI.
• Embedding provider.
• Vector database.
• Optional document processing service.
• Optional observability.

Checklist:

• Decide which documents may be indexed.
• Confirm data owners approved indexing.
• Keep provider API keys in variables.
• Restrict public access until auth is configured.
• Monitor storage growth.
• Test retrieval with non-sensitive sample data first.

AI Security Guidance​

• Do not paste production secrets into prompts.
• Do not index customer data without approval.
• Keep model provider keys in variables.
• Use separate environments for experiments.
• Avoid public access until service-level authentication is configured.
• Review service logs before sharing a new AI URL.
• Track who owns each AI stack.

AI Cost Guidance​

AI workloads can grow quickly because they combine compute, storage, and external provider spend.

Before production use:

1. Review Moltern plan limits.
2. Check the size of model runtimes.
3. Watch storage for vector databases.
4. Set spend controls in external model providers.
5. Use pay-as-you-go guardrails only with owner approval.

Example Stack Plans​

Goal	Suggested Stack
Team chat over local models	Open WebUI + Ollama
Multi-provider chat	LibreChat or LobeChat + LiteLLM
Visual agent workflows	Flowise or Langflow + LiteLLM + Qdrant
Prompt and trace review	Langfuse + model gateway
Dataset labeling	Label Studio or Argilla
Search-assisted assistant	SearXNG + agent builder + model gateway