Examples

Two runnable examples live under examples/ in the repo — both are downstream consumers, not DataCore-internal code: they only ever talk to the same REST API (and, for the Llama one, the same MCP server) any external system would use.

GitHub Portfolio (with Gemini)

Live demo · Japanese · Source: examples/github-portfolio

Registers a GitHub profile as a GITHUB_REPO resource, lets the real github-profile-scanner plugin scan it (repo metadata, README content, and the actual package.json/pom.xml/Dockerfile content for its top repos — see Building a Plugin), then:

• Hands that real data to Gemini to write portfolio copy (tagline, bio, skills, per-project highlights) — grounded only in what was actually scanned, nothing invented.
• Registers each top repo's README as its own MARKDOWN resource too, so DataCore's markdown-summarizer → vector-embedder → qdrant-register pipeline produces real SUMMARY/VECTOR artifacts for it — see Storage as a plugin.
• Builds a semantic search index from that data and serves an "Ask About My Work" search box — genuine retrieval-augmented generation over your real DataCore-processed data, not a chatbot improvising.
• Renders a two-tab minisite: Portfolio (the generated page) and DataCore Resources (every resource/artifact this run actually touched, exactly as the Core API reports it — full transparency into what ran, not a mocked demo).
• Generates both an English and a Japanese version in one run.

It's entirely optional layered on top — DataCore itself has no AI dependency; this is one example of a downstream consumer choosing to add Gemini. See the example's own README for running it locally (npm start), serving it (npm run serve), and its "Cost & abuse protection" section before making a search-enabled deployment public.

Llama + Ollama MCP Client

Source: examples/llama-mcp-client

A small demo proving DataCore's MCP server works with any tool-calling client, not just Claude — it drives a local Llama model through Ollama's OpenAI-style tools/tool_calls API, letting the model call the MCP server's tools (list_resources, search_resources, get_resource, get_artifact_content) against a real running DataCore instance to answer a question. It also demonstrates the LLM Access toggle in action: turn a resource's toggle off, ask about it by name, and the model gets a refusal instead of the resource's content — visible proof the gate actually works, not just documented behavior.

See the example's own README for setup (Ollama + a tool-calling-capable model) and usage.