Configuration

All settings live in the plugin’s settings tab, grouped into five sections.

Server

Setting Type Default Description
Port number 27182 The registry listens on 127.0.0.1 at this port (1024–65535).
Bearer token text (read-only) generated on first run Required on every request except the public catalog. Use Copy / Regenerate.
Publisher text obsidian The publisher segment of every URN (urn:air:<publisher>:…). Set a real domain you own if you ever publish externally.
Catalog name text Personal Obsidian Agentic Resource Registry Display name in the catalog’s host block.

The bind address is always 127.0.0.1 and is not user-configurable — the registry is never exposed to the network.

Skill folders

A list of folders to scan for SKILL.md files. Each input has folder autocomplete: start typing to pick a vault folder (e.g. .claude/skills), or paste an absolute path for folders outside the vault. Vault-relative picks are resolved against your vault location automatically.

  • Add folder appends a row; the trash icon removes one.
  • Watch folders for changes (off by default) auto-rescans when a SKILL.md changes — best-effort; may not fire on cloud-synced/network folders.
  • Rescan skills now re-scans the folders and rebuilds the catalog. The status line shows the last scan’s skill/error counts.

Additional resources

Manually add resources that aren’t skills — MCP server cards, A2A agent cards, nested catalogs, registries. Each needs a display name, a slug (the URN’s terminal segment), and either a URL or inline data.

Search backend

Backend Status
BM25 lexical (built-in) Default. In-process, zero download.
Local embedding server Hybrid (lexical + semantic). Uses a local embedding server you run; falls back to lexical if down.
Hosted embedding API Hybrid (lexical + semantic) via a remote provider (OpenAI/Voyage/Jina/custom); bring your own key.

The default needs no model and no network. The semantic backends add ranking quality — lexical BM25 fused (via Reciprocal Rank Fusion) with dense embeddings from an OpenAI-compatible /v1/embeddings endpoint. Nothing is bundled or downloaded by the plugin.

Local embedding server — embeddings from a server you already run (Ollama, LM Studio, llama.cpp, LocalAI, …). Configure:

  • Embedding server URL — e.g. http://localhost:11434/v1 (Ollama) or http://localhost:1234/v1 (LM Studio). Either the base /v1 or the full /v1/embeddings URL works.
  • Embedding model — e.g. nomic-embed-text.

With Ollama, a typical setup is ollama pull nomic-embed-text and leaving the defaults.

Hosted embedding API — embeddings from a remote provider. Configure:

  • Provider — openai, voyage, jina, or custom. For custom, also set an API base URL (any OpenAI-compatible gateway — Azure OpenAI, OpenRouter, a self-hosted proxy, …).
  • Model — leave blank to use the provider default (e.g. text-embedding-3-small for OpenAI).
  • API key — sent as a Bearer token; stored in plugin data, so treat it as a secret. Privacy note: your search queries and skill metadata (names, descriptions, tags) are sent to the provider to be embedded. Skill bodies are never sent.

If the embedding endpoint is unreachable, slow to start, or rejects the key, searches fall back to lexical automatically — search never breaks. Changing any backend field restarts the registry. This honors the plugin’s zero-mandatory-download principle: lexical stays the default.

Embeddings build in the background after each scan, so semantic ranking turns on a little after startup; until it’s ready, you get lexical results. On a CPU-only embedding server a large catalog (hundreds of skills) can take roughly a minute to embed the first time — a GPU-backed server, a hosted API, or a smaller model is much faster. If the embedding server starts after the plugin (or recovers from an outage), the plugin retries automatically about every 30 seconds; you can also press Reindex to pick it up immediately.

Reindex rebuilds the search index over the current catalog without rescanning your folders — useful after switching backend or to refresh a stale index. A full Rescan skills now also reindexes, so you only need Reindex when the catalog hasn’t changed.

Where settings are stored

Settings persist in the vault’s plugin data (.obsidian/plugins/agentic-resource-discovery-server/data.json). The bearer token and any API keys are stored there — treat that file as sensitive. Settings are validated on load, so a corrupt or partial file falls back to safe defaults rather than failing.