Ollama and local models

Ollama works like any other provider: on AI → Providers, add Ollama and enter its address, for example http://192.168.1.20:11434. Semestra talks to Ollama's OpenAI-compatible API under /v1, which it appends for you. No key is needed.

The private-network switchCopy Anchor Link

Ollama usually runs on localhost or a LAN address, and the Semestra server refuses to call private, loopback and link-local addresses unless SEMESTRA_ALLOW_PRIVATE_EGRESS=true. The check protects a shared server from being used to reach its own internal network.

Which address to enterCopy Anchor Link

The address is dialled by the Semestra server, not by your browser.

• Semestra in Docker, Ollama on the same host: localhost is the container itself. Use the host's LAN address, or the bundled Ollama container (below).
• Ollama on another machine: start it with OLLAMA_HOST=0.0.0.0 so it listens beyond localhost, and enter that machine's address.

The bundled Ollama containerCopy Anchor Link

The self-hosting compose file has an ai-local profile that runs Ollama next to Semestra:

docker compose --profile ai-local up -d
docker compose exec ollama ollama pull qwen3:8b
docker compose exec ollama ollama pull nomic-embed-text

Inside the compose network its address is http://ollama:11434. Add it as a provider, or make it the instance-wide provider so every user gets it without setup. Running Ollama on a GPU needs the NVIDIA Container Toolkit and a compose override; see the Ollama documentation.

CapabilitiesCopy Anchor Link

The defaults for Ollama are chat, tools, structured output and embedding. Untick tools if your model does not support tool calling, and tick vision only for a multimodal model.