A NEW WAY TO INTERACT WITH YOUR REPOSITORY

See the code.
Understand the system.

SRM builds a persistent geography of your repository—making its verified structure accessible to people and agents, and its growth traceable across map updates.

Full SRM Graph workspace showing the DesktopLab repository tree, relationship map, and inspector together — **DesktopLab / Graph**Tree, relationships and inspector in one workspace.

Scroll to explore

01 / GRAPH

Beyond the tangle.
A structure you can read.

From repository to files and relationships, context stays recognizable as you move closer.

DesktopLab Graph with source-group boundaries, visible directed routes and the open Visual key — **DesktopLab / Graph**Two marked locations lead from the full map to real directory and file views.

**DesktopLab / Graph**Two marked locations lead from the full map to real directory and file views.

IN THE DETAIL

Every element
has meaning.

Directories, files and relationships stay recognizable as you move through the map.

01 / DIRECTORYA recognizable boundary.

A source area keeps its place in the repository hierarchy.

02 / FILEAn exact resident.

Move closer to a file, then open the source behind it.

03 / RELATIONSHIPA visible connection.

Directed routes show how mapped source areas relate.

Chromium / Graph in motionScroll from repository relationships into nested source groups, then back out.

02 / CITY

The system
takes shape.

Graph foregrounds relationships. City reads the same map as a territory: a file's code lines shape its base, source bytes set its height, and language defines its silhouette. Nested directories rise as terraces above their parents, so hierarchy and scale take the lead.

DesktopLab City at closer scale, with the open Visual key showing exact source-language shapes of building footprints — **DesktopLab / City**Districts organize source areas; the open Visual key explains the language-specific building footprints.

DesktopLab / City in motionZoom, pan and tilt through the same mapped codebase without losing its layout.

03 / AT SCALE

Not a toy map.
Real codebases.

These are not static renders. Linux and Chromium are verified, navigable maps opened locally on a MacBook Pro with Apple M4 Pro and 24 GiB of unified memory—from repository scale down to exact source.

Linux v7.2 open in SRM City, zoomed until its source districts are fully colored and legible — **Linux v7.2 / City**89,392 source files · 42.3 million code lines
The complete repository becomes a territory of source districts.

Verified Chromium map open in SRM City on this Mac — **Chromium / City**169,888 source files · 33.9 million code lines
Distinct source regions remain visible at repository scale.

Linux v7.2 open in SRM Graph, with its source groups and visible relationships across the repository — **Linux v7.2 / Graph**Source groups and the relationships between them remain navigable at repository scale.

Chromium repository open in SRM Graph, showing source districts and visible magenta cross-root routes — **Chromium / Graph**Magenta routes reveal cross-root relationships in the same verified codebase.

04 / PEOPLE

Complexity grows.
Clarity remains.

Joining a codebase?

Find your bearings in the hierarchy, then follow an unfamiliar area into its files without opening dozens first.

Deep in the code?

Move between structure, relationships and exact source to investigate changes without losing the wider context.

Leading the teams?

CTOs, founders and CEOs can see how systems connect and evolve, giving cross-team discussions a shared frame of reference without reading every file.

SRM Graph with the apps source district selected and its inspector showing exact source-file and code-line counts — **DesktopLab / selected source group**The inspector reports exact source-file and code-line counts without leaving the map.

Chromium / map to sourceSelect a file and open its code inside SRM while keeping the map in view.

05 / AGENTS

A map your agent
can ask.

SRM gives an agent a read-only MCP catalog for mapped discovery, exact source and code relationships. Shell stays in the loop for the live tree, edits, builds and tests.

16read-only tools in the default workspace catalog

THE RESEARCH CATALOGREAD-ONLY / MCP

01 Orient & compare

srm_list_maps: Find ready maps and freshness.
srm_repository_status: Compare the map with checkout HEAD.
srm_repository_summary: Read source-backed counts and languages.
srm_get_changes: See map-to-HEAD and worktree path changes.

02 Discover & inspect

srm_search: Find one unknown symbol or text.
srm_search_many: Explore 2–8 independent queries together.
srm_query_graph: Follow known subjects and relationships.
srm_read_source: Read a bounded, numbered source slice.

03 Understand symbols

srm_investigate_symbol: Inspect declarations and outgoing context.
srm_find_callers: Trace direct incoming calls.
srm_get_symbol_body: Retrieve one known implementation.
srm_get_symbol_bodies: Compare 2–8 implementations.

04 Scope & prioritize

srm_explain_location: Resolve one exact source location.
srm_enrich_matches: Add context to exact source matches.
srm_inspect_scope: Inspect source genealogy and evidence.
srm_rank_refactor_candidates: Rank a declared candidate set by evidence.

Every response carries map authority and coverage. A partial result is not a confident “nothing found.”

Optional map-refresh controls 4 reviewed tools ↗

The canonical profile adds srm_map_refresh_status to read progress, srm_preview_map_refresh to review a plan, srm_execute_map_refresh to run that confirmed plan, and srm_cancel_map_refresh to stop an in-flight attempt. These are not part of the default read-only catalog.

COMPLETED / ADMITTED

Evidence before claims.

We publish measurements from completed runs that passed their declared admission gates. Each result keeps its task, delivery path and limits attached.

Host-delivered SRM context · 25 admitted MoneyChess pairs

99.2% mean answer accuracy with SRM context90.1% shell-only−79.8% total tokens across four semantic classes

Five tasks × five runs per arm; mean wall time fell from 31.6 to 15.6 seconds. This tested host-delivered context, not MCP tool choice.

SRM MCP + shell · one valid self-hosted pilot pair

−34% observed elapsed time−19.6% budgeted tokens−41.5% inferred round trips

This is an admitted efficiency observation, not a general speed or answer-quality claim: one fixed-order pair, with blind quality review not completed.

NEXT EVIDENCE

Linux long-horizon and change-replay studies are being qualified with sealed tasks and validity gates. We will publish results after their final review.

06 / PROCESS

From repository to
verified map.

A spatial map takes ten visible stages, S00–S09. SRM captures one exact source revision, inventories and analyzes it, verifies the facts, connects versions, then prepares search, visual representations and persistent geography. Activity shows each stage before the new map becomes ready.

S00–S02CaptureSource & inventory
S03–S05ModelCode facts & structure
S06–S07ConnectEvolution & search
S08–S09ProjectViews & geography

SRM Activity showing DesktopLab map creation at code analysis, with completed preparation, source capture and inventory — **DesktopLab / Activity**A real map update reporting progress from capture toward the verified spatial map.

LOCAL BY DESIGN

Your code stays
on your machine.

SRM maps and explores repositories locally. To discuss a demo, tell us a little about your codebase—never send source code.

Every product visual shown here is a real SRM capture from a MacBook Pro with Apple M4 Pro and 24 GiB of unified memory.

See the code.Understand the system.

Beyond the tangle.A structure you can read.

Every elementhas meaning.

The systemtakes shape.

Not a toy map.Real codebases.

Complexity grows.Clarity remains.