n8n-nodes-astrolabe

n8n community node for Astrolabe — sovereign EU AI for the social and solidarity economy. Astrolabe's API is OpenAI-compatible, so this node talks to api.astrolabe.chat with no code.

Installation · Credential · Operations · Development · License

Installation

Follow the n8n community nodes installation guide.

In n8n: Settings → Community Nodes → Install, then enter n8n-nodes-astrolabe.

Credential

1. Credentials → New → Astrolabe API
2. API Key: your sk-astrolabe-... key (from the developer portal)
3. Base URL: https://api.astrolabe.chat/v1 (default)

Operations

The node is organised by resource. Every knowledge-base dropdown is loaded live from your account.

AI

| Operation | Description | | --- | --- | | Message a Model | Chat completion. The model is picked from a dropdown loaded live from the Astrolabe catalogue. Returns content + reasoning_content (internal reasoning) + usage. Optionally grounds the answer on one or more knowledge bases (RAG) — the extracts used come back in astrolabe_sources. RAG supports static or agentic mode (the model searches on its own, multi-step — trace in astrolabe_steps), plus facet filter, date range, graph expansion, history window and source-link citation. | | Create an Embedding | Turns text into a vector for semantic search or RAG. The embedding model is picked from the same live catalogue. |

⚠️ astrolabe-base reasons before answering: keep Max Tokens ≥ 600 or the reply may come back empty. The node already defaults to 600.

Knowledge Base

| Operation | Endpoint | | --- | --- | | List | GET /kb | | Create | POST /kb (name, intent, type data / process, facet schema) | | Update | PATCH /kb/{slug} (rename / re-slug / change type) | | Delete | DELETE /kb/{slug} | | Search | POST /kb/{slug}/search — semantic (with a query) and/or analytic (filters, sort, group by, offset, date range, graph expansion, numeric aggregates), no model call |

Document

| Operation | Endpoint | | --- | --- | | List | GET /kb/{slug}/documents | | Add | POST /kb/{slug}/documents — text, date, external ID, source URL, facet overrides, upsert, simple / AI chunking | | Get | GET …/documents/{id} — by UUID or external ID | | Update | PATCH …/documents/{id} — title / date / external ID / source URL | | Delete | DELETE …/documents/{id} |

Chunk

| Operation | Endpoint | | --- | --- | | List | GET …/documents/{id}/chunks (decrypted) | | Add | POST …/documents/{id}/chunks (facets, date) | | Update | PATCH …/chunks/{chunkId} | | Delete | DELETE …/chunks/{chunkId} |

Notes

• List operations fan out one n8n item per result, so you can map over them directly.
• Search returns a single structured item — results (the rows) plus the analytic metadata group_by, total (exact count at the chosen granularity), truncated (page incomplete → paginate with Offset), distinct_documents (date mode) and aggregates ({field, count, sum, avg, min, max} per numeric facet). Use a Split Out node on results to fan them out. To count records, set Return = Documents and read total; to sum/average a numeric facet, set Aggregate and read aggregates.
• Analytic search: leave Query empty to list/filter only (no score, no embedding cost). Compose typed Filters (eq/neq/gt/gte/lt/lte/in/contains), exact-match Facets, Date Range, Sort and Offset freely.
• Documents and chunks are addressable by UUID or your own external_id. Adding a document with an existing external_id replaces it (idempotent upsert).
• Chunking (Add Document): Simple (paragraphs, default) or AI (intent-driven). AI chunking reveals a Chunking Model dropdown (default mistral-small) used to split the text along the base's intent — metered on your key, on top of the embedding cost. Facet Overrides (JSON) forces facet values on every chunk (takes priority over AI-extracted ones).
• Knowledge base type: a base is data (facts, default) or process (playbooks describing how to run an analysis — consulted first in agentic RAG).
• RAG on chat: select one or more Knowledge Base(s); in Options set the Knowledge Base Mode (static / agentic), Query, Facet Filter (JSON), Top K, Date Range, Graph Expansion (JSON), Cite Source Links, Search Conversation Turns and Max Tool Calls (agentic). Empty fields are omitted, so a plain chat call is unchanged.
• JSON fields (facets, filters, facet schema, facet overrides, graph expansion, aggregate): pass a JSON value as text, e.g. {"theme":"billing"}. The API accepts a JSON string or a real object.

Sovereignty

Data flows through Astrolabe's infrastructure (France / EU). Routing to a cloud model (Claude/GPT/Mistral) moves data outside the EU and is opt-in only.

Development

npm install
npm run build   # tsc + icon copy
npm run lint    # verified community node rules

To test locally, link the package into your n8n custom folder (~/.n8n/custom) with npm link.

Releasing

CI publishes to npm with a provenance statement when version in package.json is new (idempotent otherwise). To release: bump the version and merge to main. Requires the repo secret NPM_TOKEN (npm Automation token).

License

MIT