AI providers — ChatGPT, Claude, local

Which providers work?

The AI assistant in ForgeIEC is not bound to a single provider. You can freely choose:

ForgeIEC Studio automatically detects whether an endpoint speaks OpenAI or Anthropic format — you don’t have to configure that.

Connecting to ChatGPT (OpenAI)

Prerequisite: an OpenAI account with prepaid credits. The API is separate from the ChatGPT web subscription — you pay per token used, not a flat fee.

Step 1 — Create an API key

1. Sign in at https://platform.openai.com/api-keys.
2. Click Create new secret key, give it a name (e.g. ForgeIEC-WS-Joe).
3. The key starts with sk-... and is shown only once — copy it immediately into a secure place (password manager).

Step 2 — Enter it in Studio

In the AI panel, in the profile configuration:

Step 3 — Test query

Say hello.

If everything works, the assistant replies. If you see an HTTP-401 or HTTP-429 message in the output: API key wrong or rate limit exceeded.

Cost: A few cents per request. A typical programming session costs USD 0.10 - 1.00 depending on the model.

Connecting to Claude (Anthropic)

Prerequisite: Anthropic account with console access. As with OpenAI, the API is separate from the claude.ai web subscription.

Step 1 — Create an API key

1. Sign in at https://console.anthropic.com/settings/keys.
2. Click Create Key, give it a name.
3. The key starts with sk-ant-... — copy it to the password manager.

Step 2 — Enter it in Studio

ForgeIEC Studio automatically detects from the hostname api.anthropic.com that Anthropic format applies — no extra setting needed.

Step 3 — Test query

Same as above — say hello, see the reply.

Cost: Similar to OpenAI; Claude Opus is slightly pricier but often finishes long tasks faster.

Local AI (LM Studio, llama.cpp, Ollama)

When does it make sense?

• Data must not leave the network
• No per-request API costs
• You have a powerful GPU (at least 8 GB VRAM for smaller models, 24+ GB for 30B models)
• You want to try without setting up an account

Variant A — LM Studio (easiest)

1. Download + install LM Studio from https://lmstudio.ai.
2. Pick a model in the LM Studio browser, e.g. qwen3-coder-30b-a3b-instruct for coding tasks.
3. In the Developer tab, start a local server — default at http://localhost:1234.
4. In the ForgeIEC editor:

   Field	Value
   API	http://localhost:1234/v1
   API Key	(leave empty or anything)
   Model	as shown in LM Studio

Variant B — llama.cpp server

llama-server -m model.gguf --port 1234 — same API URL as LM Studio.

Variant C — Ollama

1. Start ollama serve, pull a model (ollama pull qwen2.5-coder).
2. Address: http://localhost:11434/v1 (OpenAI-compatible endpoint).

Model recommendation for PLC coding (as of 2026-05):

• For local operation, at least a code-specialised 30B model (e.g. Qwen-3-Coder-30B-A3B, DeepSeek-Coder-V2-Lite)
• Smaller models (7B-14B) make many IEC-ST syntax errors and often loop

Which provider for which task?

Multiple providers in parallel?

Yes. You can set a different provider per persona:

• Blacksmith Master → gpt-4o (fast, good coding)
• Reviewer → claude-opus-4-7 (thorough)
• Monitor → small local model (safe, free)

Switch in the AI panel between the tabs — settings are saved per persona.

Security + privacy

Cloud providers:

• Your requests + the ST code snippets the assistant sends pass through the internet to the respective provider.
• If that is not acceptable for your project → local AI.
• Never share API keys with colleagues or check them into Git.

Local AI:

• Data does not leave your machine.
• The API key is optional / irrelevant (no real server checks it).

In all cases:

• The ForgeIEC editor itself stores no requests or replies in the cloud.
• Chat history is saved locally under ~/.config/ForgeIEC/Chats/<persona>.jsonl.
• The audit log at ~/.config/ForgeIEC/mcp_audit.log records every write action with a timestamp — regardless of provider.

Next

• Personas — characters
• Daily operation — input + emergency stop
• Security model
• Back to the AI overview