Model server
Loaded in memory
Installed models
| Name | Family | Params | Quant | Size | Status |
|---|
Pull a model
This box has 4 CPU cores and ~5 GB free for models — anything above ~4 GB (e.g. 8B models) will swap and be painfully slow. Good Meta picks: llama3.2:3b (installed), llama3.2:1b (2× faster, less capable), llama-guard3:1b (safety classifier).
API keys
External sites and systems authenticate with these. Secrets are shown once at creation and stored hashed.
| Name | Key | Origins | Rate/min | Daily tokens | Today | 24h reqs | Last used | Status |
|---|
Daily usage (14 days)
| Day | Key | Requests | Tokens | Avg latency |
|---|
Recent requests
| Time | Key | Kind | Model | Tokens | Latency | Status | Origin / IP | Error |
|---|
Connect external sites & systems
Base URL: . Authenticate with Authorization: Bearer <key>. The API is OpenAI-compatible, so any OpenAI SDK or tool works by pointing its base URL here.
1. Website chat widget (any site, one script tag)
Create a key with the site's hostname in Allowed origins and a system prompt describing what it should answer. Then paste before </body>:
2. Simple question → answer (curl / any language)
3. OpenAI-compatible chat (Python)
4. OpenAI-compatible chat (Node.js / JavaScript)
5. Streaming with the OpenAI SDK
6. Embeddings (semantic search)
7. No-code tools (Zapier, n8n, Make, WordPress AI plugins, LangChain…)
Anywhere a tool asks for an OpenAI API key and lets you set a custom base URL: set base URL to , API key to your dsk_… key, and model to . Tools that hard-code api.openai.com cannot be pointed here.
Rate limits per key are enforced (HTTP 429 with retry_after). The model runs on CPU: expect ~1–3 s for short answers, up to ~30 s for long ones, and requests queue when more than 2 run at once.
Defaults
How the platform itself uses this
DigiSqull's own ai-service and course-service talk to the same Ollama directly (env AI_PROVIDER=ollama, AI_MODEL). Changing the default model here affects the public API and widget; to change the platform's model edit AI_MODEL in infrastructure/.env and redeploy those two services.