Your codebase, in perspective Preview 0.4.2

Turn your repository
into a map you can read.

Move from the big picture to the source behind it.
Explore files, symbols and relationships in one connected workspace.

Continue in the app to choose your analysis scope.

No account or API key. Local file analysis stays in your browser.

skylense / autoresearchCurated example · actual workspace
Skylense's actual workspace showing the curated autoresearch experiment flow and connected component details
FILES → CLASSES → FUNCTIONSCONNECTED SOURCE EVIDENCEWEB · CLI · MCPSIX WORKSPACE THEMES
01 / BRING YOUR SOURCE

A link. A folder. A file.

Read a public repository, choose local files or open a saved map.

02 / SEE WHAT WAS READ

Coverage comes first.

Review analyzed files, skipped content and unresolved relationships.

03 / EXPLORE THE CONNECTIONS

Keep every layer in view.

Follow relationships, add route waypoints and open the source evidence.

From overview to understanding

Find your place.
Then follow your curiosity.

Navigate a generated source map with the same hierarchy, relationship lenses, multi-point paths and source previews as the example workspaces.

A real Skylense hierarchy view with expanded modules and component source details
01

Keep the bigger picture.

Expand folders and files into source modules, classes, functions and methods. Highlight peers, dim unrelated modules and keep your parent context.

02

Find a place to start.

Choose a source guide by its entry name, module and relationship purpose. See its objects highlighted in the same tree, follow connections, then return to your previous view.

03

Read the evidence.

Inspect line-numbered source excerpts and the evidence behind a relationship. GitHub source links stay pinned to the analyzed revision.

A map that moves with you

Follow an idea
through the system.

Readable right-angle connections and guided flows make complex relationships easier to explain. Pause motion whenever you need.

Recorded Skylense animation showing relationship direction in the autoresearch map

Actual app recording. Animated connections indicate direction in a curated map, not live execution or measured traffic.

Bring your own source

Your project.
The complete workbench.

Open a folder, a public GitHub repository, or a webpage. Review what was read, then explore the same layers, paths and source previews.

Complete indexing is the default. 24 bundled syntax grammars, cached unchanged files, and project-aware reference resolution feed the same workbench. Large maps unfold by module or page; the full graph remains searchable. Unsupported syntax and uncertain dispatch stay explicit.

Start with something real

Three AI codebases.
Ready to explore.

Hand-organized architecture stories at fixed source revisions. Use these to discover the workbench, then bring your own project.

These curated examples include annotated concepts and guided flows. Automatic source maps show the files, symbols and static relationships found in your selected source; they do not invent the same semantic narrative.

Make room for your way of working

Six themes.
Your kind of focus.

Warm paper, cool porcelain, or a quieter night shift. Switch themes without losing your place.

Skylense in the Midnight theme

Midnight — deep blue surfaces and an ice-blue focus.

Beyond the browser

The same map.
Where you work.

Keep your terminal open. Give your coding agent a structured view of the architecture. Both read the same model as the Web app.

Terminal explorer

Explore a generated map from your terminal: expand the tree, search symbols and read the same relationships with your keyboard.

Nine MCP tools

Analyze a source, export its model, then search, inspect, traverse and open the full visual workspace. Connect a client that supports local stdio MCP.

Read the connection guide
Terminal / after installing the toolkit

# Open your project in the full Web app

$ skylense open ./project

# Save a map for your tools

$ skylense analyze ./project \
    --output project-map.json

# Give your agent access to this folder

$ skylense config --root ./project

Get from curious to exploring

Open a new perspective.

Try it now, or keep an offline copy for your next deep dive.

  1. 01 Download the Web ZIP
  2. 02 Extract the folder
  3. 03 Open standalone.html
Get the Agent & Terminal toolkit (Node.js 22+) →
Can I explore my own repository?

Yes. Enter a public GitHub repository above, or choose a local folder, files, webpage or text in the app. Complete eligible-source indexing is the default, with 24 bundled syntax grammars, incremental file reuse and project-aware references. JavaScript/TypeScript also uses the compiler symbol table; Julia preserves possible method targets as marked candidates. The coverage report separates source collection, syntax errors, unresolved semantics and preview excerpts. Large maps expand progressively without deleting analysis data.

How do source guides relate to the tree?

The tree shows where code lives. Source guides group connected objects to read together, with short entry names, module context, categories and exact relationship counts. Browse the current module or all guides; selections stay synchronized with the graph and tree. These are static reading neighborhoods with possible branches, not runtime traces. The full graph remains available for path search and upstream/downstream exploration.

What if a URL cannot be read?

Check that the repository is public and the URL points to a repository or subtree. GitHub rate limits and browser CORS rules can prevent remote reading. You can select a local checkout instead, or use skylense open ./project with the downloadable toolkit. Read the source guide.

Is this a native macOS app?

This release is a browser app with offline HTML, plus a local Node.js toolkit. A native macOS installer is not available yet.

Is Skylense open source?

Skylense is available as a downloadable application preview. Source licensing and commercial rights remain reserved. The attributed examples, bundled parsers and compiler retain their upstream licenses. Read the preview notice.

What happens to an imported model?

Models stay in the current browser session. Syntax facts and verified public GitHub file bytes are cached locally to speed up repeat analysis; selected source files are not uploaded. The local MCP process reads requested public URLs and source files inside an explicitly configured root. Connected agents receive query results according to their own data policies. Read the data boundaries.

What language is the interface?

The preview's interface is primarily Chinese, with source symbols in their original language. This site and the getting-started guides are available in English; a Chinese README is also provided.