Research environment for people and agents

Asterelle

Pick up any research project without spending hours reconstructing the context.

Plain Markdown Local-first Human ratification
Explore the product Try it with my research

A personal research superapp · Keep your Markdown and Git repositories · macOS, Windows, and Linux

Not a new file format or another chat history. Asterelle keeps project context, AI work, and human decisions together.

Changes and pending decisions from every project in Asterelle Observatory
From interrupted context to the next decisionObserve · Review · Continue

00 · The workbench

See all your research from one screen.

Notes, code, equations, Git, task boards, and research agents remain connected inside a visible project boundary.

Research notes and relationships in Asterelle Constellation

01 · How it works

From opening your research folder to keeping an approved record.

There is no proprietary import step. Asterelle reads the workspace you already have and connects it to the next decision.

One morning with Asterelle

Open an analysis project you paused last week. Observatory gathers the recent changes and overnight verification results. Inspect the Git evidence, delegate the follow-up to a Mate, and promote only the result you trust into the project record.

01 / Open

Open your existing research folder.

Markdown, code, data, and Git repositories stay where they are. Asterelle reads their project boundaries and relationships.

02 / Observe

See what needs your judgment.

Observatory brings project changes, blocked work, overnight results, and pending decisions together with their evidence.

03 / Work

Work directly or delegate.

Continue in files, notebooks, and terminals, or give Mates a bounded task inside the active project.

04 / Review

Review the result with its evidence.

Inspect the proposed change, execution history, file and Git evidence, authorship, and verification state in one review item.

05 / Ratify

Promote only what you trust.

Only approved candidates enter the canonical research state. Rejections and deferrals remain part of the decision history.

02 · One research system

Not a collection of tools. A connected research flow.

  1. ConstellationSee how ideas and evidence connect.

    Map relationships among papers, concepts, methods, claims, and projects from links already present in Markdown. Indexes are rebuildable; source notes remain plain files.

  2. Files & WorkbenchKeep reading and execution close together.

    Scientific Markdown, code, notebooks, terminals, Git, and task boards follow the same project boundary. Scope stays visible in both local and SSH workspaces.

  3. MatesResearch collaborators with visible roles and relationships.

    Project agents prepare, review, and verify through explicit delegation paths. Progress and execution history persist beyond the conversation.

  4. ObservatoryBring today's decisions into one place.

    Signals, overnight results, and connection candidates from every project arrive with evidence. No result enters through a silent automatic edit.

Your research stays yours.

  • Canonical notes remain plain Markdown inside your workspace.
  • Existing Git repositories stay usable outside Asterelle.
  • Search and graph indexes are derived and rebuildable.
  • Provider and locality settings control model calls.

03 · One environment, many models

Use the right model for each part of the research.

Bring the model access you already use. Asterelle routes configured roles through one project context, then applies the same evidence and review rules to every result.

Claude CodeCodexLocal modelsConfigured providers
One research contextProject scope · execution history · evidence · human review

01 / BRING

Use what you already have.

Connect existing Claude Code or Codex access and configured local or remote providers instead of committing the whole workflow to one model vendor.

02 / ASSIGN

Assign models by role.

Use different models for coordination, coding, reading, summarization, overnight preparation, and independent verification.

03 / CONTINUE

Keep one project memory.

Changing the model does not discard the task, execution record, evidence references, or the project boundary.

04 / CONTROL

Control cost and locality.

Keep routine or sensitive preparation local when configured, and reserve stronger remote models for the work that needs them.

Model availability and data handling follow the providers and locality settings configured by the researcher. Asterelle does not silently promote a fallback result into the canonical record.

04 · How it earns trust

The final judgment always belongs to the researcher.

  1. 01 Agents propose. Researchers promote.Changes to research state pass through explicit gates. Agents never hold final authority.
  2. 02 Separate authorship from verification.Independent criteria and roles distinguish producing an artifact from judging it.
  3. 03 The canonical record stays inside the project.Markdown is canonical. Graph and search indexes are rebuildable derived state.
  4. 04 Unattended work is optional.Night automation runs only queued work within configured limits and returns evidence for review.

05 · Private beta

Try it in your own research flow.

Asterelle is being tested with a small group of individual researchers in real research workspaces. Tell us about your environment and we will send a suitable desktop build with a guided starting point.

  • macOS · Windows · Linux
  • Uses your existing Claude Code or Codex access
  • Does not modify system Python
  1. Tell us your field and setup
  2. We check workspace fit
  3. Receive a suitable build
  4. Start with one real project
Get early access

06 · FAQ

What to know before entrusting your research context.

How is this different from adding an AI plugin to Obsidian?
Obsidian connects notes. Asterelle carries those relationships into project-scoped code, task boards, persistent agent runs, and explicit research gates.
Can agents change the research record on their own?
No. Agent output remains a candidate until it passes explicit confirmation that the agent cannot issue for itself.
What information leaves my computer?
Model calls follow the provider and context boundaries configured for each workflow. Research state stays in the workspace, while deployment settings control models and locality for night work.
What do I need to run it?
You need a research workspace and access to Claude Code or Codex. Asterelle provisions its runtime in a product-specific directory without modifying system Python.
Can I keep using my existing Obsidian vault?
Yes. Asterelle indexes Markdown in place and interprets wiki-links as graph relationships. Files remain readable in other editors, and derived indexes can be rebuilt.
When can I use it?
Asterelle is currently in a directly distributed private beta. Tell us about your environment and we will contact you when a suitable build is ready.

Still have questions? Email us directly.

07 · Where Asterelle is going

Starting personal. Growing into a deeper research environment.

These are directions under exploration, not current features. Personal, local-first workspaces will remain the foundation.

Exploring

Research Library

Connect paper collection, metadata, citation, annotation, and a focused reading mode to research notes and writing.

Exploring

Deeper evidence

Capture provenance among claims, data, executions, and documents more precisely, with stronger reproducibility checks.

Exploring

Lab Workspace

Explore a collaboration layer for shared projects, reviews, explicit roles, and auditable decisions without giving up personal workspaces.

The roadmap is not a release commitment. Scope will follow researcher interviews and real-world validation.