Skip to content

Recipes#

Each recipe builds one real thing from start to finish: the models it needs, the knowledge base, the steps in the console, with astra and with the API, what you should see, what it holds on your machines, and how to clean up. Start with the one closest to what you are doing.

  • Your company handbook, searchable with citations

    PDF and Word files in a folder on one of your machines, made a drive, indexed, searched with the page or heading of every answer, and measured with an evaluation set.

    Uses: a drive on one machine, BGE-M3 and a reranker on CPUs, astra metis create, search, evaluate.

  • A personal notes base on a CPU-only machine

    Markdown notes on a small machine with no GPU, added from your laptop and from your phone, with the CPU defaults for the two models.

    Uses: a base with no drive, astra metis add, the app, CPU deployments.

  • Search your knowledge from Claude Code

    One command installs the Astralyx skill; Claude Code searches your bases, cites the documents it used, and saves a note only after your yes.

    Uses: astra connect claude-code, astra metis search --json, astra metis note --yes.

  • Chat with your documents in Eos

    A chat model that searches a base as a tool and cites what it found, and saves a note to the base when you ask — after you confirm.

    Uses: a function as a tool, save_to_knowledge, the Playground and the app's chat.

  • A memory for your assistant

    Remember this conversation from the app or the Playground, dated facts reviewed and corrected in Metis ▸ Memory, and an assistant that answers from them.

    Uses: a memory, Qwen3 4B as its extractor on a CPU, astra metis memory.

  • Measure and improve search quality

    An evaluation set, recall@k and MRR, the same set with and without the reranker, the three modes side by side in Search, and what to change when questions are missed.

    Uses: Metis ▸ Evaluations, Metis ▸ Search, evaluate with mode and rerank.

  • Keep a base in sync with a data source

    A bucket listed every hour as a drive with revisions; each new revision indexed incrementally, what changed visible pass by pass.

    Uses: a data source, a drive from it, revisions, astra astraeus drives refresh.

How the recipes are written#

  • Every step shows each way to do it, in the order Console, CLI, API. The CLI is astra metis (CLI and MCP tools), signed in with astra use <org>/<workspace>; a few things are in the console and the API only (a base with no drive, a memory, comparing settings in an evaluation), and each recipe says so.
  • Names you choose are written as the recipe chooses them (handbook, bge-m3); <org>/<workspace> stands for yours.

  • API calls go to https://api.astralyx.cloud/v1 with an API token scoped to your workspace, made under Account → API tokens (scope One workspace); below, ast_pat_… stands for it:

    $ export ASTRALYX_API=https://api.astralyx.cloud/v1
$ export ASTRALYX_TOKEN=ast_pat_…

  • You need a workspace with a cluster, a machine with a few free CPU cores and a data location (Machines → the machine → Data location → Confirm) — a knowledge base keeps its index there — and the editor or admin role.