Skip to main content
entire-graph is an installable plugin for the Entire CLI that gives your coding agent a map of your codebase. It helps your agent find the right code, understand how it connects, and see what a change could affect before it starts editing. Install entire-graph before running entire graph:
Then run entire-graph commands with:
Run entire graph help to see every command in your installed version, and entire graph COMMAND --help for command-specific flags. Pass --repo . when you run these outside an Entire session.

graph setup

Write .entire/agent-guide.md, add managed blocks to AGENTS.md and CLAUDE.md, and register the Claude Code SubagentStart hook in the per-user .claude/settings.local.json (skip the hook with --no-claude-hook). Replaces the removed init-agents command.
See Install and Enable entire-graph for the complete setup workflow.

graph agent-guide

Print the coding-agent guide that setup installs.

graph index

Build a committed-tree cache before running a batch of --head queries. The default profile is full.

graph capabilities

Report supported languages, parsing coverage, relation types, and features as JSON.

graph query

Return ranked source regions for a plain-language description of a task or bug.
You can also pass the text as the final argument instead of using --query:
Query reads the working tree by default, so results include uncommitted changes. Use --head to search the committed tree instead. Results fit within a byte budget. Top results include full function bodies, while later results may contain only locations. Use --max-context-bytes to change the total. Each result includes a signals array that explains why it ranked, such as path, body, symbol-name, or graph:callers.

graph impact

Show callers, callees, type consumers, data flows, related files, and same-container symbols for one symbol.
Impact reads the working tree by default. Use --head to inspect the committed tree instead.
Text output begins with Index: cache-hit or Index: cache-miss.

graph neighbors

Show direct relationships for one symbol, filtered by relation and direction.
Neighbors reads the working tree by default. Use --head to inspect the committed tree instead. Text output begins with Index: cache-hit or Index: cache-miss.

graph def

Show a symbol’s declaration. For a type, include its fields, methods, and implementations.
This command reads the working tree by default. Use --head to inspect the committed tree instead.

graph explain

Read a build or test failure from standard input and append the declarations of the symbols it names.
Explain reads the working tree by default, so its declarations match the code that produced the failure. Use --head to inspect the committed tree instead.

graph commit

List entity-level changes in a commit relative to its first parent, including added, removed, renamed, signature-changed, and body-changed symbols.
Each change includes an estimated dependent count.

graph diff

List entity-level changes between two Git references. analyze is an alias for this command.

graph checkpoint

Analyze the commit associated with an Entire checkpoint.

graph snapshot

Export the complete graph as NDJSON, including files, external endpoints, symbols, relations, and a summary.
Snapshot reads the committed tree by default. Use --worktree to include uncommitted changes. It also supports --format compact-ndjson.

graph symbols

Export symbol records as NDJSON.
Symbols reads the committed tree by default. Use --worktree to include uncommitted changes. Use query or def for targeted lookups because this command does not filter by name.

graph edges

Export relation records as NDJSON. Filter the output with --to, --from, and --relation.
Edges reads the committed tree by default. Use --worktree to include uncommitted changes.

graph snapshot-query

Query a saved compact snapshot without rebuilding the graph.

graph doctor

Report the resolved repository, Entire environment variables, plugin data directory access, and whether network access is disabled.
doctor --assert is a preflight. It parses a command line against your installed binary and tells you whether it would be rejected, without running anything:

graph health

Report how completely entire-graph indexed the committed tree, including when the result is healthy.
The report lists flagged source files and their percentage of the total, the 5% threshold at which coverage counts as degraded, a breakdown by language and diagnostic type, and the affected files with parser diagnostic locations. Documentation, configuration, and data files are not counted as source files. Files skipped on purpose, such as oversized or minified files, are listed separately. A parser diagnostic means entire-graph could not fully analyze a file. It does not mean the source code is invalid. Health always uses the committed tree at HEAD and the full profile, so it does not include uncommitted changes. It reuses a matching index or builds one first. Use doctor to check the environment and health to check index coverage.

graph version

Print the installed entire-graph plugin version. Add --json to include the provider name.

graph stats

Compare entire-graph usage with grep and file reads in local coding-agent transcripts, including an estimated token saving.
stats reads local coding-agent session transcripts from disk. It is for humans, not part of the agent workflow, and the installed guide tells agents not to run it unless asked.

Continue with

Use entire-graph with Your Coding Agent

Learn what to ask and what happens after you prompt your agent.

Install and Enable entire-graph

Install the plugin and turn it on for a repository.

entire-graph Troubleshooting

Fix install, adoption, and cache problems.

entire-graph Overview

Understand what entire-graph is for.

entire-graph Repository

Browse the source and the full reference docs.