🚀 EDDI 6.5.0

19 LLM providers, knowledge bases that crawl your site and read your files, and conversations that follow compatible agent versionsWhat's new

Managed Cloud Get Started
All models
Mistral AIRuns locally

Ministral 3 (3B, 8B, 14B)

Small open-weight models in three sizes, each with base, instruct and reasoning variants and image understanding, for local and edge use.

Run It in EDDIVendor documentation ↗

Strengths

  • Built for local deployment
  • Reasoning variants in every size
  • Image understanding
  • Apache 2.0 license

Best For

  • Edge and on-premises deployment
  • Low-cost, high-volume tasks
  • Local agents

Run It in EDDI

The fastest route is the Manager: open the agent wizard or the Platform Operator and pick the provider and model. To configure it yourself, follow these steps for the provider you want.

Mistral AI · type: "mistral"

Model ids: ministral-14b-2512

1. Store the credential

Add the provider's API key to the Secrets Vault in the Manager, or with PUT /secretstore/secrets/default/mistral-key, under the name used below. Configurations reference it as ${vault:…}, so the key never appears in plain text.

${vault:mistral-key}

2. Configure the LLM task

Use this as the langchain.json of the agent's workflow, or paste the parameters into the LLM editor in the Manager.

{
  "tasks": [
    {
      "actions": [
        "send_message"
      ],
      "id": "chat",
      "type": "mistral",
      "parameters": {
        "apiKey": "${vault:mistral-key}",
        "modelName": "ministral-14b-2512",
        "systemMessage": "You are a helpful assistant",
        "addToOutput": "true"
      }
    }
  ]
}

Or create the whole agent in one call

setup_agent writes the rules, LLM configuration and workflow, then deploys the agent. Call it over MCP or POST /administration/agents/setup.

setup_agent(
  agentName: "My ministral-14b-2512 agent",
  systemPrompt: "You are a helpful assistant.",
  provider: "mistral",
  model: "ministral-14b-2512",
  apiKey: "${vault:mistral-key}"
)

Notes for EDDI

  • A good first tier in a model cascade: fast and inexpensive, escalating to a larger model only when confidence is low.

Related Models

Mistral AIMistral Small 4An open-weight mixture-of-experts model (about 6B active parameters) that combines instruct, reasoning and coding, with image input.MicrosoftPhi-4-miniMicrosoft's 3.8B open model for memory-constrained, latency-sensitive work and maths and logic reasoning.GoogleGemma (Gemma 4, Gemma 3)Google's open-weight family, from edge-sized models to a 31B dense and a 26B mixture-of-experts model in Gemma 4.

Sources

The facts on this page come from the vendor's documentation, checked on 2026-09-30. Models change quickly, so confirm limits with the vendor before you rely on them. Prices are not listed here: see the vendor's pricing page.

Vendor documentation ↗ · Pricing ↗