Connecting an Artica appliance to DeepSeek

Connect the Artica AI assistant of an appliance to the hosted DeepSeek API: benefits, concepts, use cases and step-by-step setup.

Back to Artica AI main section

This page connects the Artica AI assistant of an appliance to the DeepSeek API.
The assistant then writes its answers and its investigations with a hosted DeepSeek model.
Nothing has to be installed besides the Artica AI service itself.
For the principles of Artica AI, read the Artica AI page first.

Functional benefits

Answers in plain language.
The assistant explains what it reads on the appliance with real sentences.
Without a model, it only prints the raw results of its own checks.
Automatic investigations.
When an incident opens, the assistant chains read-only checks by itself.
It then writes a diagnosis and, when useful, a remediation plan.
No hardware to buy.
The model runs at DeepSeek.
The appliance needs no GPU and no extra memory.
Ready in two minutes.
Install the service, choose DeepSeek, paste the key, apply.
Low cost.
The deepseek-flash model is fast and cheap.
It fits the short investigations of the assistant well.
Same safety rules.
The model only writes text.
Every change still goes through the policy, an approval and a verified rollback.

Concepts

Term	Meaning
Artica AI service	The daemon that runs on the appliance. It reads the state of the appliance, keeps the audit log and applies the policy.
Language model	The remote server that writes the text of the answers. Here, the DeepSeek API.
Endpoint	The address of an OpenAI-compatible API. For DeepSeek: `https://api.deepseek.com/v1`.
API key	The secret that identifies your DeepSeek account. It starts with `sk-`. The console never displays it again once saved.
Deterministic mode	The mode used when no model is configured. Every answer comes from the local rules engine. Nothing leaves the appliance.
Remote mode	The mode used with a hosted API such as DeepSeek. Questions and observations are sent over HTTPS. Secrets are masked before they are sent.
Offline mode	A switch that stops all calls to the model. The settings and the key are kept. The assistant falls back to deterministic mode.
Agent	The loop that lets the model run read-only checks one after the other. It is bounded by budgets: 8 turns, 12 checks and 180 seconds by default.

How it works

The appliance keeps control at every step.
The model never reaches the system directly.

An administrator asks a question, or an incident opens.
Artica AI chooses read-only checks from its catalog and runs them on the appliance.
The results are cleaned: passwords, keys and tokens are masked.
Artica AI sends the question and the cleaned results to DeepSeek over HTTPS.
DeepSeek may ask for more checks.
Artica AI runs only the read-only checks that its catalog allows.
DeepSeek returns a diagnosis.
Artica AI displays it with the evidence it collected.
When a fix is needed, the assistant proposes a plan.
The plan waits for the approval of an administrator, unless the Autopilot allows it.

The model only proposes.
The policy decides.
Artica AI never runs a free shell command.

If DeepSeek fails or answers too late, the assistant still answers.
It shows its own deterministic summary and the error of the model.
It never invents a diagnosis.

Use cases

Situation	What DeepSeek brings
First evaluation of Artica AI	Test the full assistant at once, with no local AI server.
Small site or single appliance	A local AI appliance would be oversized. A hosted model needs no hardware at all.
Daily operations questions	"Is the proxy service healthy?", "Why is the DNS cache slow?". The answer is a short readable diagnosis.
Incident follow-up at night	The assistant investigates each new incident by itself. The diagnosis is ready when the administrator arrives.
Appliance without spare resources	A proxy or a mail gateway with little memory cannot host a local model. DeepSeek does the heavy work outside.

When the data must stay inside your network, use an Artica AI appliance instead.
It serves a local model to your other appliances: see Connecting an Artica appliance to the Artica AI appliance.

Prerequisites

An Artica appliance with the Artica AI service available in Your system.
A DeepSeek account with credit, and an API key created in the DeepSeek platform.
Outgoing HTTPS (TCP 443) from the appliance to api.deepseek.com.
An administrator account on the Web console.

Step 1: install the Artica AI service

Open Your system, then Artica AI.

When the service is not installed yet, the page shows what it brings.
Click Installation.

The installation takes a few seconds.
The left menu then shows the pages of the assistant, and the top bar shows the Ask ArticaAI button.
The service starts in deterministic mode: the card Engine and language model reads Deterministic mode (no language model).

Step 2: connect DeepSeek

In the card Engine and language model, click Parameters.

In Provider, choose DeepSeek.
The endpoint https://api.deepseek.com/v1 is filled in.
In Model, choose DeepSeek V4.1 Flash (deepseek-flash).
Keep Answer Timeout at 300 seconds.
Leave Offline Mode off.
In API Key, choose Replace the key.
Paste your DeepSeek key in New API Key.
Click Apply.

Before the first save, the line under the model may read The server did not return its model list: HTTP 401.
This is expected.
DeepSeek lists its models only to a known key, and the key is not saved yet.

Saving restarts the Artica AI service.
For a few seconds the page may show The artica-ai service does not answer.
Click Refresh.

Step 3: check the connection

The card now reads Remote language model, with the badge REMOTE.
The model is deepseek-flash and the key shows as Configured.

Open Parameters again.
With the saved key, DeepSeek now returns its list: the line reads Models offered by the server: 2.
This proves that the appliance reaches DeepSeek and that the key is accepted.
Close the window without applying.

Step 4: ask a first question

Click Ask ArticaAI in the top bar.
A terminal opens.
Type a question and press Enter.

The answer starts with the status, the risk level and the session number.
The Diagnosis section is the text written by DeepSeek.
Type /verbose to display the hidden sections: checks, evidence and plan.
Type /help to list the other commands.
The Detach button opens the terminal in its own browser window.

Choosing the model

Model	When to use it
`deepseek-flash`	Recommended. Fast and cheap. It completes the investigations of the agent well within their 180 seconds.
`deepseek-v4-pro`	Deeper reasoning, but slower and more expensive. Raise the investigation timeout of the agent if you choose it.

Cost and data

Each question sends one or more requests to DeepSeek.
Each new incident starts an investigation, which also calls DeepSeek.
The card Assistant agent limits them: 6 investigations per hour by default.
The data sent is the question and the results of read-only checks.
Passwords, keys and tokens are masked before sending.
Host names, addresses and service states are sent.
DeepSeek is a hosted service outside your network.
Check that this fits your data policy.
To stop all calls at once without losing the settings, turn on Offline Mode.

Troubleshooting

Symptom	Cause and fix
HTTP 401 under the model, after the key was saved	DeepSeek refuses the key. Create a new key in the DeepSeek platform, then use Replace the key.
The answers end with a model error	DeepSeek is unreachable, too slow, or the account has no credit left. Check the outgoing HTTPS flow to `api.deepseek.com` and the balance of the account.
The card still reads Deterministic mode	The endpoint is empty or Offline Mode is on. Open Parameters and check both fields.
The artica-ai service does not answer	The service is restarting after a save. Wait a few seconds and click Refresh.