Skip to main content
Atlas maps your workspace into a queryable knowledge graph: functions, types, classes, and files, connected by calls, imports, and inherits edges, and clustered into subsystems. Instead of grepping, you (and the AI agents on your branch node) can ask the graph directly: how does auth work, what breaks if I change this, what’s the shortest path from the HTTP handler to the database layer. The graph is built entirely locally from the code — no network access, no LLM, and no API key required. It is derived state of one branch’s working tree: every branch node builds and serves its own graph, so switching branches means looking at that branch’s graph.

​
Building the graph

The build is incremental — files are re-extracted only when their content changes — so after the first build, rebuilds are fast. Once a workspace has built its graph, the branch node keeps it fresh automatically: file changes you make in the editor trigger a debounced incremental rebuild in the background. Optional enrichment flags:

​
Asking questions

query accepts --depth, --budget, and repeatable --context filters to restrict traversal to particular edge kinds (calls only, imports only, and so on); when you don’t pass a filter, one is inferred from the question itself.

​
Impact analysis

Two commands answer “what does this change touch”:
affected walks the reverse edges from a symbol and reports everything that depends on it — with the call sites, not just the definitions. impact diffs your branch against a base ref, maps the changed lines to symbols, and reports the blast radius and the subsystems it touches: a review map for the branch.

​
Understanding the shape of the codebase

​
The Atlas view in the editor

The editor renders the same graph as an interactive view alongside your code. Large graphs open as a subsystem overview — one node per community — rather than a hairball; double-click a subsystem to drill into it, and running a query shows exactly the answer’s subgraph. Smaller graphs render in full.
  • Click a symbol to see its explanation — identity and connections.
  • Double-click a symbol to open its file at the defining line.
  • Search for a symbol to jump to it across views.
Nodes are colored by subsystem and sized by how connected they are, and edge styling reflects the extractor’s confidence in each relationship. The generated wiki is also readable inside the editor — an Atlas Wiki tab in the semantic search panel renders one article per subsystem, with links that navigate within the view. If a branch node hasn’t built its graph yet, the Atlas view says so and offers to build it.

​
Work-memory: the graph learns your branch

Query outcomes can be recorded into the branch’s committed history:
Saved outcomes live in the committed .reasonos/ tree, so they travel with the branch through git — the sources that answered questions well rank higher in future queries, and dead ends rank lower, for everyone on the branch and for the agents working on it.

​
Querying other repositories

Every read command accepts --workspace to query a graph you haven’t cloned:
The reference can be an @org/project[@branch] slug resolved through the control plane, a local workspace path, or a node URL.