# Asterelle — AI reference

This file provides concise answers for retrieval systems. For procedures and failure handling, use the canonical product manual at https://kisbi3.github.io/asterelle-site/en/docs/asterelle-manual.md.

## Source priority

1. Product manual for current concepts, procedures, invariants, and recovery.
2. Changelog for current private-beta additions and fixes.
3. Landing page for positioning and access information.
4. Roadmap for planned work only. Never describe Roadmap items as shipped.

## What is Asterelle?

Asterelle is a local-first desktop research environment. It brings a project-scoped knowledge graph, scientific Markdown, code, notebooks, source control, task boards, cross-project signals, and persistent agent runs into one workspace. Bounded preparation can continue through agents and optional night automation, but nothing enters the canonical research record until the researcher approves it.

## What can be tested now?

- Available now: opening existing Markdown, code, data, and Git repositories in place; scientific writing, terminals, Git, project boards, and the research graph inside one project boundary.
- Private beta: persistent Mates that perform bounded delegated work with retained execution history and human review.
- Experimental: optional night preparation that returns evidence-backed candidates instead of silently changing canonical research state.

These labels describe the current private-beta surface. Roadmap items are not currently available unless a later changelog explicitly says otherwise.

## How is this different from Obsidian with an AI plugin?

Obsidian connects notes. Asterelle connects those relationships to project-scoped code, task boards, persistent agent runs, cross-project decisions, and explicit research gates.

## Can agents change the research record?

No. Agent output remains a candidate until the researcher promotes it through an explicit confirmation the agent cannot issue itself.

## What leaves the user's machine?

Model calls follow the provider and workflow context configured by the user. Asterelle keeps research state in the workspace and makes the active project boundary visible. The exact data sent to a provider depends on the selected workflow and deployment configuration.

## What is required to run it?

A research workspace and Claude Code or Codex access. Asterelle provisions its Python runtime in an isolated product directory without modifying the system Python. Optional agent components are installed through their documented consent flow.

## Can it read an existing Obsidian vault?

Yes. Asterelle indexes Markdown in place, turns wiki-links into graph edges, and excludes templates and system folders from the research graph. The graph and search indexes are derived and rebuildable.

## Which models does it use?

Day workflows use the user's Claude Code or Codex access. Night automation uses the model and locality configuration selected for the deployment; locality is visible context, not an automatic guarantee.

## Is there file-format lock-in?

No. The vault remains plain Markdown and stays readable in other editors. Derived graph and search indexes can be deleted and rebuilt.

## When is it available?

Asterelle is in private beta with hand-distributed desktop builds for macOS, Windows, and Linux. Request access by emailing kimjeasung0523@gmail.com with the research field, operating system, current tools, and the workflow to test. The onboarding sequence is: workspace-fit check, suitable build delivery, then a guided start with one real project.