Skip to content

v2.26.0

Latest

Choose a tag to compare

@kOaDT kOaDT released this 18 Nov 21:29
· 3 commits to main since this release

Breaking Changes

  • The AI assistant no longer asks for a Mistral API key in the browser. The model is now configured on the server with LLM_BASE_URL, LLM_MODEL and LLM_API_KEY. By default it calls gemma4:e4b on a local Ollama (http://localhost:11434/v1 with npm, http://ollama:11434/v1 in the Docker image). @kOaDT
    • Who is impacted: anyone playing the prompt injection or malicious MCP server challenges, in particular players who saved a Mistral key in the chat UI, and anyone running the published image with docker run. That image now expects a container named ollama on the same Docker network.
    • Action required:
      • npm: install Ollama, then run ollama pull gemma4:e4b. Nothing else to configure.
      • Docker Compose: run npm run docker:up:ollama, then npm run docker:ollama -- pull gemma4:e4b.
      • Keep using Mistral: set LLM_BASE_URL="https://api.mistral.ai/v1", LLM_MODEL="mistral-small-2603" and LLM_API_KEY in .env, or pass them with -e to docker run.
    • Full instructions are in the new AI assistant setup guide.

What's New

  • Run the AI challenges offline, or on any OpenAI-compatible endpoint. The assistant works with any provider whose model supports tool calling. A local model needs no account and no API key, and needs no internet once the model is downloaded. npm run setup and npx create-oss-store add the three LLM_* variables to new .env files as comments. @kOaDT
  • Ollama in Docker. A new docker-compose.ollama.yml runs Ollama next to the lab, with no published port and with models kept in a persistent volume. npm run docker:up:ollama starts both containers, and npm run docker:ollama runs the Ollama CLI inside the container. @kOaDT
  • In-app setup help. When the assistant can't answer, the chat shows what went wrong: the endpoint is unreachable, the model is missing, the credentials were rejected, or the rate limit was hit. It shows the configured model and host, and links to the setup guide. With a local Ollama, it also gives the ollama pull command to run. @kOaDT
  • Startup check. An LLM_BASE_URL that is not an http(s) URL now stops the app at startup with an explicit error, instead of failing on the first chat message. @kOaDT

Improvements

  • New AI assistant setup guide for npm, Docker Compose, docker run and hosted providers. It covers model choice, troubleshooting, and a table of tested models showing why gemma4:e4b is the default. The README, EDUCATORS.md, the AI walkthroughs, the About page and the comparison table now point to it. @kOaDT
  • The README is shorter and keeps what players and educators read. Project structure and test commands now live in CONTRIBUTING.md. @kOaDT

Maintenance / Chore

  • The AI assistant API tests and CI now use a local OpenAI-compatible stub instead of a live Mistral key. Contributors running npm run test:api start the server with LLM_BASE_URL=http://127.0.0.1:11435/v1 LLM_MODEL=stub LLM_API_KEY=stub-key. @kOaDT
  • CI validates the Compose files, Ollama overlay included, before building the Docker image. @kOaDT