100% local · graph db + contract standard

wiregraph

Two engines, one graph. A local call graph lets Claude ask instead of re-read — ~half the tokens. A contract standard it enforces across your repos carries that reach to the seams between services — where other code-graph tools stop at one repo's edge.

Install → Why it reaches further →

~½

the bytes a full read pulls

1×

index, then forget

bytes uploaded

GRAPH.DB

call · within a compartment contract · across compartments

The problem → the fix

Stop paying to re-read the whole tree.

A code task shouldn't cost a full grep-and-read of your repo. wiregraph turns "search everything" into "ask the graph."

🐢 Without wiregraph

01 grep the whole tree
02 read whole files, top to bottom
03 change it — full token cost

tokens readbaseline

100%

⚡ With wiregraph

01 query the graph
02 read just what matters
03 change it from the full picture — ~half the cost

tokens readabout half less

~47%

The differentiator

Works with your contract architecture.

Within a compartment, calls resolve by name. Across compartments there's usually no call at all — services agree on a shared shape instead. wiregraph treats that shape as a standard: it detects the seam a call graph can't see, so Claude traces a cross-compartment change from the graph instead of re-reading both services — the same token savings, now across the boundary.

→
A route, a queue topic, an env var, or an imported package — when a token appears in 2+ repos, wiregraph mints a Contract node.
→
Producer and consumer both link to it with REFERENCES edges, so Claude can trace producer → contract → consumer across the boundary.
→
Rename one side and the seam is right there — no grep, no guessing which of your services just broke.
→
Repos indexed separately? /wiregraph-link joins two graphs so a shared wire lights up across the boundary — symmetric, one db per graph, no shared code required.

What is contract architecture? → Guide & commands →

What Claude can do

Seven tools. One graph.

You don't call these — Claude does, automatically, on every task.

🔍

find_symbol

answers where something is defined.

📄

get_source

answers one symbol's body — not the whole file.

⬆️

trace_callers

answers who calls it — the whole tree, one query.

⬇️

trace_callees

answers what it calls — the whole tree, one query.

🔗

trace_contract

answers how code connects across compartments, via a contract.

🧭

path_between

answers the wiring between any two symbols.

🗄️

query_sql

answers anything else — read-only SQL over the graph.

✨

graph_status · update_graph

keep it fresh — automatically.

// set it up once and forget it — the graph stays fresh as you edit.

// joining two separately-indexed repos? /wiregraph-link makes trace_contract cross the seam between them — see the guide.

Install

Three lines. No build step.

Public repo, install straight from Claude Code. Native bits are prebuilt and vendored for Linux / macOS / Windows.

claude-code

/plugin marketplace add kaleLetendre/wiregraph

/plugin install wiregraph@wiregraph

/reload-plugins

/wiregraph-init

# → indexes every repo under this workspace, then keeps itself fresh

Run /wiregraph-init once per workspace. It finds every repo under the root, indexes them into one graph, and re-indexes as you edit — git-worktree-safe, so adding a worktree, rebasing, or amending never forces a rebuild or leaves the graph silently stale. Then just put Claude to work.