docscode intelligence

Know what breaks before you touch it.

Code Map scans your repo into a local symbol graph, then computes a compiler-verified blast radius for any symbol — so you and your agents change the right code with the right context, and nothing leaves your machine.

overview

Code Map (the MAP view) scans your local codebase into a graph of files, symbols, imports and dependency edges, persisted on disk as JSON. Search for a symbol and see exactly what depends on it and what it uses, then compute a compiler-verified blast radius before you change it. It exists so you — and coding agents — can find the smallest relevant slice of a repo without sending the whole thing anywhere or scanning it blindly.

  • Graph built and stored on-device under .13x/map/ — local only, never uploaded
  • Compiler-verified dependents via a local TS-resolver sidecar, not keyword guessing
  • Rename-safe / NOT rename-safe verdict with dynamic sites flagged explicitly
  • No embeddings, no cloud indexing, no external AI calls, no telemetry
  • Agent map packets reference files, symbols and neighbors — never file contents
  • Debounced 500ms watcher with incremental splice for clean in-root edits
why it matters

Compiler-verified blast radius

Selecting a symbol calls a local TS-resolver sidecar for verified direct dependents, then BFS-expands transitive dependents over compiler-resolved edges only. The view never falls back to keyword guessing for the dependency picture.

Honest about what it can't see

Results carry a rename-safe / NOT rename-safe verdict and flag explicit 'dynamic' sites where static analysis can't see through, like dynamic dispatch. The radius tells you where it's incomplete instead of pretending it's complete.

100% local, no telemetry

The whole graph is built and stored on your machine under .13x/map/. No embeddings, no cloud indexing, no external AI calls — and the agent-context prompt itself instructs agents not to send code or telemetry anywhere.

Feeds agents the smallest relevant slice

Turn a blast radius into a budget-scoped Context Packet or SEND TO AGENT, so agents read the right files first instead of grepping the whole repo. AGENT rig slots can also auto-inject a compact map packet on launch.

Live, incremental updates

A debounced 500ms filesystem watcher re-indexes on .ts/.tsx/tsconfig.json changes. Clean in-root edits splice just that file's symbols and edges; adds, removes and renames trigger a full rescan, then the UI refreshes automatically.

Deterministic and cache-backed

Ranking, search and the BFS estimate are deterministic with stable tie-breaks. Verified blast results are cached keyed by symbol and graph generation, then cleared on every rescan — instant on repeat, never stale.

how it works
01

Scan the map

Pick a project path and click SCAN MAP. The indexer walks the repo, and if a tsconfig.json exists it calls the ts-resolver sidecar to add compiler-resolved edges, then persists the graph to .13x/map/codegraph.json.

02

Find a symbol

Type into the find-symbol box to search by name or path (exact > prefix > contains), or browse the MOST DEPENDED-ON ranking when the box is empty.

03

Inspect the blast radius

Select a symbol to trigger verified blast-radius computation. The inspector shows USED BY and USES neighbors, direct/transitive/depth counts, a rename-safety verdict, and any dynamic sites.

04

Build a context packet

Pick a budget — tiny, small, balanced or deep — and click CREATE PACKET to generate task-specific Markdown with files/symbols/neighbors and a token-reduction metric. Preview, copy, or SEND TO AGENT.

05

Keep it live

The watcher re-indexes on relevant changes and emits code_map_updated; the MAP view reloads automatically when the event root matches your active project.

faq

Does any of my code leave my machine?

No. The graph is built on-device and written under .13x/map/ inside the scanned project — local only, never uploaded. There are no embeddings, no cloud indexing, no external AI calls and no telemetry. Agent map packets reference ranked files, symbols and dependency neighbors, but contain no file contents.

What does 'compiler-verified blast radius' actually mean?

Selecting a symbol calls a local TS-resolver sidecar that runs against your tsconfig.json to get verified direct dependents, then BFS-expands transitive dependents over compiler-resolved edges only. It carries a rename-safe verdict and flags dynamic sites where static analysis can't see through, so the radius is honest about where it's incomplete.

Which languages does it support?

The indexer extracts symbols from TypeScript/TSX and Rust, plus package.json scripts and Cargo.toml metadata. Resolved dependency edges and the verified blast radius are TypeScript/TSX-centric and require the TS-resolver sidecar and a tsconfig.json. Rust symbols are indexed, but the verified-blast path is driven by the TS sidecar.

Is this a visual node graph?

Not in the MAP view — it renders a symbol list plus an inspector. The interactive dependency-graph rendering lives in the separate Symbol Graph view. Note also that symbol extraction is regex-based, not a full AST parse, so it's optimized for finding the right slice rather than perfect compiler analysis.

get started

one project. many shells. zero noise.

13x is local-first and free of telemetry — your code, commands and shells never leave the machine. Join the waitlist for the next build.