KiroGraph

Semantic code knowledge graph for Kiro: fewer tool calls, instant symbol lookups, 100% local.

Inspired by CodeGraph by colbymchenry for Claude Code, rebuilt natively for Kiro's MCP and hooks system.

Why KiroGraph?

When you ask Kiro to work on a complex task, it explores your codebase using file reads, grep, and glob searches. Every one of those is a tool call, and tool calls consume context and slow things down.

KiroGraph gives Kiro a semantic knowledge graph that's pre-indexed and always up to date. Instead of scanning files to understand your code, Kiro queries the graph instantly: symbol relationships, call graphs, type hierarchies, impact radius — all in a single MCP tool call.

The result is fewer tool calls, less context used, and faster responses on complex tasks.

What Gets Indexed?

KiroGraph uses tree-sitter to parse your source files into an AST and extract:

• Nodes — functions, methods, classes, interfaces, types, enums, variables, constants, routes, components, and more (24 node kinds total)
• Edges — calls, imports, exports, extends, implements, contains, references, instantiates, overrides, decorates, type_of, returns

Everything is stored in a local SQLite database (.kirograph/kirograph.db). Nothing leaves your machine. No API keys. No external services.

The index is kept fresh automatically via Kiro hooks — no background watcher process needed.

How Indexing Works

Indexing has three layers: structural (always on), semantic (opt-in), and architecture (opt-in).

Structural indexing

tree-sitter parses every source file into an AST. Nodes and edges are extracted and written to kirograph.db. This is what powers all graph traversal tools (kirograph_callers, kirograph_impact, kirograph_path, etc.) and exact/FTS symbol search.

This layer has no extra dependencies and runs on every kirograph index or kirograph sync.

Semantic indexing (opt-in)

When enableEmbeddings: true is set, KiroGraph additionally generates 768-dimensional vector embeddings for every embeddable symbol (function, method, class, interface, type_alias, component, module) using the nomic-ai/nomic-embed-text-v1.5 model (~130MB, downloaded once to ~/.kirograph/models/).

These embeddings power natural-language search in kirograph_context and act as a fallback in kirograph_search. The embeddings are stored in the semantic engine of your choice:

Engine	Store	Search type	Extra deps
cosine (default)	kirograph.db (vectors table)	Exact cosine, linear scan	none
sqlite-vec	.kirograph/vec.db	ANN (approximate), sub-linear	better-sqlite3, sqlite-vec (native)
orama	.kirograph/orama.json	Hybrid (full-text + vector)	@orama/orama, @orama/plugin-data-persistence
pglite	.kirograph/pglite/	Hybrid (full-text + vector), exact	@electric-sql/pglite (WASM)
lancedb	.kirograph/lancedb/	ANN (approximate), sub-linear	@lancedb/lancedb (pure JS)
qdrant	.kirograph/qdrant/	ANN (HNSW), sub-linear	qdrant-local (embedded binary)
typesense	.kirograph/typesense/	ANN (HNSW), sub-linear	typesense (auto-downloaded binary)

Each engine owns its embedding store exclusively — nothing is written to the SQLite vectors table when a non-cosine engine is active. If an engine's optional dependency is not installed, KiroGraph silently falls back to cosine.

Enable and configure via kirograph install (interactive arrow-key menu) or directly in .kirograph/config.json:

{
  "enableEmbeddings": true,
  "semanticEngine": "pglite"
}

Architecture analysis (opt-in)

When enableArchitecture: true is set, KiroGraph detects the high-level structure of your project — packages and architectural layers — and computes coupling metrics between them. Results are stored in arch_* tables inside kirograph.db and exposed via dedicated MCP tools and CLI commands.

Enable via kirograph install or directly in .kirograph/config.json:

{
  "enableArchitecture": true
}

See the Architecture Analysis section below for full details.

Installation

From npm (not yet available on npm registry)

npm install -g kirograph

From source

git clone https://github.com/davide-desio-eleva/kirograph.git
cd kirograph
npm install
npm run build
sudo npm install -g .

After building, the kirograph and kg commands are available globally.

Verify

kirograph --version

Uninstallation

Remove from a project

kirograph uninit [path]     # Remove .kirograph/, hooks, and steering file from the project
kirograph uninit --force    # Skip confirmation prompt

This removes:

• .kirograph/ — index database, snapshots, and export directory
• .kiro/hooks/kirograph-*.json — all KiroGraph hooks
• .kiro/steering/kirograph.md — the steering file
• .kiro/agents/kirograph.json — the CLI agent config

Remove the CLI globally

If installed from npm:

npm uninstall -g kirograph

If installed from source:

cd kirograph
npm uninstall -g .

Quick Start

# In your project:
kirograph install    # wire up MCP + hooks + steering + CLI agent in .kiro/

Restart Kiro IDE, or switch to the kirograph agent in Kiro CLI. It will now use KiroGraph tools automatically.

Or using the short alias:

kg install

How It Works

┌─────────────────────────────────────────┐
│                  Kiro                   │
│                                         │
│  "Fix the auth bug"                     │
│           │                             │
│           ▼                             │
│  kirograph_context("auth bug")          │
│           │                             │
└───────────┼─────────────────────────────┘
            ▼
┌───────────────────────────────────────────┐
│         KiroGraph MCP Server              │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐   │
│  │  search  │ │ callers  │ │ context  │   │
│  └────┬─────┘ └────┬─────┘ └────┬─────┘   │
│       └────────────┼────────────┘         │
│         SQLite Graph DB (.kirograph/)     │
└───────────────────────────────────────────┘

Kiro hooks mark the index dirty on every file save or create, then flush on agent idle — batching changes efficiently with no overhead during active editing.

Using with Kiro

kirograph install sets up four things in your Kiro workspace — all coexist, so you can switch between IDE and CLI freely:

MCP Server (.kiro/settings/mcp.json)

Registers the KiroGraph MCP server. Used by both the IDE and the CLI agent:

{
  "mcpServers": {
    "kirograph": {
      "command": "kirograph",
      "args": ["serve", "--mcp"],
      "autoApprove": [
        "kirograph_search", "kirograph_context", "kirograph_callers",
        "kirograph_callees", "kirograph_impact", "kirograph_node",
        "kirograph_status", "kirograph_files", "kirograph_dead_code",
        "kirograph_circular_deps", "kirograph_path", "kirograph_type_hierarchy",
        "kirograph_architecture", "kirograph_coupling", "kirograph_package"
      ]
    }
  }
}

IDE Auto-Sync Hooks (.kiro/hooks/)

Four hooks keep the index fresh automatically in the Kiro IDE:

Hook	Event	Action
kirograph-mark-dirty-on-save.json	fileEdited	kirograph mark-dirty
kirograph-mark-dirty-on-create.json	fileCreated	kirograph mark-dirty
kirograph-sync-on-delete.json	fileDeleted	kirograph sync-if-dirty
kirograph-sync-if-dirty.json	agentStop	kirograph sync-if-dirty --quiet

File changes are batched: saves and creates write a dirty marker; the actual sync runs when the agent stops. Deletes sync immediately. This means no overhead during active editing.

Hooks fire for: .ts, .tsx, .js, .jsx, .py, .go, .rs, .java, .cs, .rb, .php, .swift, .kt, .dart

CLI Agent Config (.kiro/agents/kirograph.json)

A custom agent for Kiro CLI that wires up the MCP server, inlines the steering instructions as a prompt, and handles sync in the CLI's own hook format. The CLI has no file-watch events, so syncing is handled at session boundaries instead:

Hook	Event	Action
agentSpawn	Agent starts	kirograph sync-if-dirty --quiet — catches edits made between sessions
userPromptSubmit	Each prompt	kirograph sync-if-dirty --quiet — keeps graph fresh within a session
stop	End of each turn	kirograph sync-if-dirty --quiet — deferred flush, mirrors IDE agentStop

Use it with:

kiro-cli --agent kirograph

Or swap to it inside an active session:

/agent swap kirograph

Note: restart kiro-cli after running kirograph install for the agent to be picked up.

Steering File (.kiro/steering/kirograph.md)

Teaches the Kiro IDE to prefer graph tools over file scanning when .kirograph/ exists. The CLI agent has the same instructions inlined directly in its prompt field.

MCP Tools

All tools are auto-approved and available to Kiro once installed.

kirograph_context

Comprehensive context for a task or feature — often sufficient alone without additional tool calls.

Parameter	Type	Default	Description
task	string	required	Task, bug, or feature description
maxNodes	number	20	Max symbols to include
includeCode	boolean	true	Include code snippets
projectPath	string	cwd	Project root path

How it works: Extracts symbol tokens from the task description (CamelCase, snake_case, SCREAMING_SNAKE, dot.notation) → runs exact name lookup + FTS + vector search against the active semantic engine → resolves imports to their definitions → expands through the graph to related symbols → returns entry points, related nodes, edges, and code snippets. This is the only tool that uses the vector engine on every call.

kirograph_search

Quick symbol search by name. Returns locations only, no code.

Parameter	Type	Default	Description
query	string	required	Symbol name or partial name
kind	string	—	Filter: function, method, class, interface, type_alias, variable, route, component
limit	number	10	Max results (1–100)
projectPath	string	cwd	Project root path

How it works: Exact name match → SQLite FTS → LIKE fallback → vector search only if all three return nothing. Pure graph database lookup in the common case; vector engine only as a last resort.

kirograph_callers

Find all functions/methods that call a specific symbol.

Parameter	Type	Default	Description
symbol	string	required	Symbol name
limit	number	20	Max results (1–100)
projectPath	string	cwd	Project root path

How it works: BFS traversal of incoming call edges in the graph database. No vector engine involved.

kirograph_callees

Find all functions/methods that a specific symbol calls.

Parameter	Type	Default	Description
symbol	string	required	Symbol name
limit	number	20	Max results (1–100)
projectPath	string	cwd	Project root path

How it works: BFS traversal of outgoing call edges in the graph database. No vector engine involved.

kirograph_impact

Analyze what code would be affected by changing a symbol. Use before making changes.

Parameter	Type	Default	Description
symbol	string	required	Symbol name
depth	number	2	Traversal depth
projectPath	string	cwd	Project root path

How it works: BFS traversal of all incoming edges (call, import, reference, etc.) up to the specified depth. No vector engine involved.

kirograph_node

Get details about a specific symbol, optionally including source code.

Parameter	Type	Default	Description
symbol	string	required	Symbol name
includeCode	boolean	false	Include source code
projectPath	string	cwd	Project root path

Returns: kind, name, qualified name, file location, signature, docstring, and optionally source code.

How it works: Single row lookup by symbol name in the graph database. If includeCode is true, reads the relevant lines directly from the source file on disk. No vector engine involved.

kirograph_type_hierarchy

Traverse the type hierarchy of a class or interface.

Parameter	Type	Default	Description
symbol	string	required	Class or interface name
direction	string	both	up (base types), down (derived types), both
projectPath	string	cwd	Project root path

How it works: Recursive traversal of extends and implements edges in the graph database. No vector engine involved.

kirograph_path

Find the shortest path between two symbols in the dependency graph.

Parameter	Type	Default	Description
from	string	required	Source symbol name
to	string	required	Target symbol name
projectPath	string	cwd	Project root path

How it works: BFS shortest-path search across all edge types in the graph database. No vector engine involved.

kirograph_dead_code

Find symbols with no incoming references (potential dead code). Only unexported symbols are considered.

Parameter	Type	Default	Description
limit	number	50	Max results (1–100)
projectPath	string	cwd	Project root path

How it works: Queries the graph database for nodes with zero incoming edges, filtered to non-exported symbols. No vector engine involved.

kirograph_circular_deps

Find circular import dependencies in the codebase.

Parameter	Type	Default	Description
projectPath	string	cwd	Project root path

How it works: Tarjan's strongly connected components algorithm over import edges in the graph database. No vector engine involved.

kirograph_files

List the indexed file structure with filtering and format options.

Parameter	Type	Default	Description
filterPath	string	—	Filter by directory prefix (e.g., src/)
pattern	string	—	Filter by glob pattern (e.g., **/*.ts)
maxDepth	number	—	Limit tree depth
format	string	tree	tree, flat, or grouped
includeMetadata	boolean	true	Include language and symbol counts
projectPath	string	cwd	Project root path

How it works: Reads file records from the graph database and builds a tree structure in memory. Filtering is applied before tree construction. No vector engine involved.

kirograph_status

Check index health and statistics: files indexed, symbol count, edge count, breakdown by kind and language, frameworks detected, database size, and semantic search status.

Parameter	Type	Default	Description
projectPath	string	cwd	Project root path

How it works: Reads aggregate counts from the graph database + calls count() on the active vector engine to report embedding coverage. No graph traversal, no vector search.

kirograph_architecture (requires enableArchitecture: true)

Get the full architecture overview: detected packages, layers, and the dependency graph between them.

Parameter	Type	Default	Description
projectPath	string	cwd	Project root path

Returns: packages (with source, language, version, external deps, file membership), layers (with file counts and detection patterns), package dependency edges, layer dependency edges, and per-file package/layer assignments.

How it works: Reads the arch_* tables populated during the last kirograph index run. Returns nothing useful if architecture analysis was not enabled at index time.

kirograph_coupling (requires enableArchitecture: true)

Get coupling metrics for all packages or a specific one.

Parameter	Type	Default	Description
packageId	string	—	Package ID (e.g. pkg:npm:src/auth). Omit for all packages.
projectPath	string	cwd	Project root path

Returns per-package: Ca (afferent — how many other packages depend on this one), Ce (efferent — how many packages this one depends on), and instability (Ce / (Ca + Ce), 0 = maximally stable, 1 = maximally unstable). When packageId is given, also returns the full list of incoming and outgoing package dependencies.

kirograph_package (requires enableArchitecture: true)

Inspect the files and dependencies of a specific package.

Parameter	Type	Default	Description
packageId	string	required	Package ID (e.g. pkg:npm:src/auth)
projectPath	string	cwd	Project root path

Returns: package metadata, all files assigned to the package, packages it depends on (with import counts), and packages that depend on it.

kirograph_hotspots

Find the most-connected symbols by total edge degree (incoming + outgoing). Excludes structural contains edges.

Parameter	Type	Default	Description
limit	number	20	Max results (1–100)
projectPath	string	cwd	Project root path

Returns each symbol with total degree, in-degree, and out-degree. Useful for identifying core abstractions and high blast-radius code before making changes.

kirograph_surprising

Find non-obvious cross-file connections: direct edges between symbols in structurally distant files.

Parameter	Type	Default	Description
limit	number	20	Max results (1–100)
projectPath	string	cwd	Project root path

How it works: Queries all cross-file edges (excluding contains and import). Scores each by path distance between source and target files × edge-kind weight (calls=1.0, references=0.8, type_of=0.7, etc.). Returns the highest-scoring unique pairs — the ones that represent the most unexpected coupling in the codebase.

kirograph_diff

Compare the current graph state against a saved snapshot. Shows added/removed symbols and edges.

Parameter	Type	Default	Description
snapshot	string	latest	Snapshot label. Omit to use the most recent saved snapshot.
projectPath	string	cwd	Project root path

Use kirograph snapshot save (CLI) to save a snapshot before a refactor or PR. Run kirograph_diff after to see what changed structurally.

CLI Reference

Setup

kirograph install                 # Wire up MCP + hooks + steering in .kiro/
kirograph init [path]             # Initialize .kirograph/ in a project
kirograph init --index            # Initialize and index immediately
kirograph uninit [path]           # Remove .kirograph/, hooks, and steering file
kirograph uninit --force          # Skip confirmation prompt

Indexing

kirograph index [path]            # Full re-index of the project
kirograph index --force           # Force re-index all files (ignore hash cache)
kirograph sync [path]             # Incremental sync of changed files
kirograph sync --files a.ts b.ts  # Sync specific files only
kirograph sync-if-dirty [path]    # Sync only if a dirty marker is present
kirograph mark-dirty [path]       # Write a dirty marker for deferred sync

Status & Maintenance

kirograph status [path]           # Show index stats (files, symbols, edges, frameworks)
kirograph unlock [path]           # Force-release a stale lock file

Search & Exploration

kirograph query <term>                    # Search symbols by name
kirograph query <term> --kind class       # Filter by kind
kirograph query <term> --limit 20         # Limit results (default: 10)

Supported kinds: function, method, class, struct, interface, trait, protocol, enum, type_alias, property, field, variable, constant, enum_member, parameter, import, export, route, component, file, module, namespace

File Structure

kirograph files [path]                     # Show indexed file tree
kirograph files --format flat              # Flat list of all files
kirograph files --format grouped           # Files grouped by language
kirograph files --filter src/components    # Filter by directory prefix
kirograph files --pattern "**/*.test.ts"   # Filter by glob pattern
kirograph files --max-depth 2              # Limit tree depth
kirograph files --no-metadata              # Hide language/symbol counts
kirograph files --json                     # Output as JSON

Context Building

kirograph context "fix checkout bug"
kirograph context "add user authentication" --format json
kirograph context "refactor payment service" --max-nodes 30
kirograph context "validate token" --no-code

Extracts symbol tokens from the task description (CamelCase, snake_case, SCREAMING_SNAKE, dot.notation), finds relevant entry points, expands through the graph, and outputs structured markdown or JSON.

Affected Tests

Find test files that depend on changed source files — useful in CI or pre-commit hooks.

kirograph affected src/utils.ts src/api.ts           # Pass files as arguments
git diff --name-only | kirograph affected --stdin     # Pipe from git diff
kirograph affected --stdin --json < changed.txt       # JSON output
kirograph affected src/auth.ts --filter "e2e/**"      # Custom test file glob
kirograph affected src/lib.ts --depth 3 --quiet       # Paths only, shallow traversal

Option	Description	Default
--stdin	Read file list from stdin, one per line	false
-d, --depth <n>	Max dependency traversal depth	5
-f, --filter <glob>	Custom glob to identify test files	auto-detect
-j, --json	Output as JSON	false
-q, --quiet	Output file paths only	false
-p, --path <path>	Project path	cwd

Example CI integration:

#!/usr/bin/env bash
AFFECTED=$(git diff --name-only HEAD | kirograph affected --stdin --quiet)
if [ -n "$AFFECTED" ]; then
  npx vitest run $AFFECTED
fi

🪨 Caveman Mode 🪨

Caveman mode compresses the agent's communication style, cutting token usage on responses without affecting tool calls or code output. Inspired by caveman 🪨 by JuliusBrussee.

Why it's useful: KiroGraph's graph tools return compact, structured data. The bottleneck in long coding sessions isn't the tool calls — it's the verbose prose the agent wraps around them. Caveman mode strips that overhead so you get the signal without the filler. The rules are injected at session start via the steering file (IDE) and the inline agent prompt (kiro-cli), so they're always in context with no extra tool calls.

Four levels:

Mode	Style
off	Normal responses (default)
lite	Compact, no filler, full sentences
full	Fragments, no articles, short synonyms
ultra	Maximum compression, abbreviations, → for causality

kirograph caveman lite    # compact, still readable
kirograph caveman full    # fragments, no articles
kirograph caveman ultra   # maximum compression
kirograph caveman off     # back to normal
kirograph caveman         # show current mode

Set during kirograph install (interactive arrow-key menu) or any time after. Takes effect on the next agent session.

Caveman mode never touches code blocks, file paths, URLs, or technical terms — only prose.

Auto-clarity exceptions: the agent temporarily reverts to normal prose for security warnings, confirmations of irreversible actions (delete, overwrite, force-push), and multi-step sequences where fragment order could cause misunderstanding. Compressed style resumes immediately after.

Architecture Analysis (requires enableArchitecture: true)

Visualize the detected package graph, architectural layers, and package dependencies.

kirograph architecture [path]              # Show packages + layers + all deps
kirograph architecture --packages          # Show packages section only
kirograph architecture --layers            # Show layers section only
kirograph architecture --format json       # JSON output

Output includes:

• Each detected package with its source (manifest or directory), language, version, and declared external deps
• Package-to-package dependency edges with import counts
• Detected layers (api, service, data, ui, shared) with file counts
• Layer-to-layer dependency edges

Package Inspection (requires enableArchitecture: true)

Drill into a single package: metadata, coupling metrics, dependencies, and files.

kirograph package <name>                   # Inspect a package by name or path fragment
kirograph package auth                     # Partial match accepted (e.g. matches "pkg:npm:src/auth")
kirograph package src/auth --no-files      # Omit file list
kirograph package auth --format json       # JSON output

Shows package source (manifest or directory), language, version, manifest path, coupling metrics (Ca/Ce/instability), outgoing dependencies, incoming dependents, declared external deps, and the full list of files belonging to the package.

Coupling Metrics (requires enableArchitecture: true)

Inspect coupling health across your package graph.

kirograph coupling [path]                  # All packages, sorted by instability
kirograph coupling --sort ca               # Sort by afferent coupling (most depended-on first)
kirograph coupling --sort ce               # Sort by efferent coupling (most dependent first)
kirograph coupling --sort name             # Sort alphabetically
kirograph coupling --package auth          # Detail view for a single package
kirograph coupling --format json           # JSON output

The table shows each package with:

• Ca — afferent coupling: how many packages depend on this one (higher = more stable)
• Ce — efferent coupling: how many packages this one depends on (higher = more unstable)
• Instability — Ce / (Ca + Ce), rendered as a color-coded bar: green (stable) → yellow (neutral) → red (unstable)

The --package detail view shows who depends on this package and what it depends on, with import counts for each relationship.

Hotspots

Find the most-connected symbols in the codebase by total edge degree (incoming + outgoing, excluding structural contains edges). Useful for identifying core abstractions, load-bearing code, or high blast-radius change points.

kirograph hotspots [path]             # Top 20 most-connected symbols
kirograph hotspots --limit 10         # Limit results
kirograph hotspots --format json      # JSON output

Output shows each symbol with an inline bar chart, total degree, and in/out breakdown.

Surprising Connections

Find non-obvious cross-file connections: direct edges (calls, references, etc.) between symbols in structurally distant parts of the codebase. High-score pairs indicate unexpected coupling worth investigating.

kirograph surprising [path]           # Top 20 surprising connections
kirograph surprising --limit 10       # Limit results
kirograph surprising --format json    # JSON output

Score = path distance between files × edge-kind weight (calls=1.0, references=0.8, type_of=0.7, etc.).

Snapshots & Diff

Save lightweight graph snapshots and compare them to track structural changes over time — useful before/after refactors, or in CI to audit what a PR added or removed.

kirograph snapshot save [label]       # Save current graph state with optional label
kirograph snapshot save pre-refactor  # Named snapshot
kirograph snapshot list               # List all saved snapshots
kirograph snapshot diff               # Diff current graph vs latest snapshot
kirograph snapshot diff pre-refactor  # Diff current graph vs named snapshot
kirograph snapshot diff --format full # Show full added/removed symbol lists
kirograph snapshot diff --format json # JSON output

Snapshots are stored in .kirograph/snapshots/ as JSON and include all node IDs and edge tuples. The diff is computed as a set operation — O(n) regardless of codebase size.

The kirograph_diff MCP tool exposes the same capability to the agent: compare the current graph against the latest (or a named) snapshot without leaving the conversation.

Dead Code

Find unexported symbols with zero incoming references — candidates for removal.

kirograph dead-code [path]            # List dead code grouped by file
kirograph dead-code --limit 20        # Limit results
kirograph dead-code --format json     # JSON output

Only unexported symbols are considered, since exported symbols may be used by consumers outside the indexed project.

Path

Find the shortest connection between any two symbols, traversing all edge types in both directions.

kirograph path <from> <to>            # Find path between two symbols
kirograph path LoginController Pool   # Example: how are these connected?
kirograph path --format json          # JSON output

The command resolves symbol names using the same fuzzy search as kirograph query, preferring real symbol kinds (class, function, method…) over import/file nodes. The result shows each hop with file and line.

Graph Export

Export the full graph as an interactive dashboard — three files served from a local directory, no server required, works offline.

kirograph export build [path]             # Generate .kirograph/export/{index.html,app.css,app.js}
kirograph export start [path]             # Generate and open in browser
kirograph export build -o /tmp/myexport  # Custom output directory
kirograph export build --include-contains # Include structural contains edges (adds noise, off by default)

Output lands in .kirograph/export/ by default. Open index.html in any browser.

Graph & navigation

• Color-coded nodes by kind (class, function, method, component…) with size proportional to degree
• Directed edges with kind labels; dashed lines for imports and references
• Click a node to zoom in and inspect it — kind, file, line, degree, signature, and a copy button for the file reference
• Click two nodes to instantly find and highlight the shortest path between them, with detail cards for both endpoints
• History — ‹ › navigation through previously inspected nodes
• Keyboard shortcuts — f to fit the graph, Esc to exit focus or path mode

Controls

Button	What it does
⊞ Fit	Fit the entire graph to the viewport
⚡ Physics	Toggle the force-directed layout
⛶ Fullscreen	Collapse the side panel for maximum graph space
📷 PNG	Save the current view as an image
◎ Focus	Show only the selected node and its direct neighbors
⟶ Path	Find the shortest path between two nodes
⬡ Cluster	Group nodes by directory; click a cluster to expand it
🌡 Heat	Color nodes by how recently their file was modified
📊 Charts	Open the analytics panel

Search

Type to search by name, qualified name, or file path. Matching nodes are highlighted and the viewport fits to them.

Legend & filters

• Node kind filter — Legend tab; click any kind to hide or show all nodes of that type
• Edge kind filter — Legend tab; click any edge kind to hide or show edges of that type
• Degree slider — Filters tab; hide nodes below N connections to surface the most-connected symbols

Minimap

An overview of the full graph is always visible in the bottom-left corner. Click anywhere on it to pan the main graph.

Right-click menu

Right-click any node to focus its neighbors, start a path from it, copy its ID or file path, or highlight all nodes of the same kind.

Analytics charts

The 📊 Charts button opens a panel with three charts:

Chart	What it shows
Bar	The 15 most-connected symbols
Donut	How node kinds are distributed across the codebase
Line	How many symbols have each connection count — reveals the overall connectivity shape of the graph

Dashboard

When semanticEngine is set to qdrant or typesense, use these commands to manage the background server and its dashboard UI.

kirograph dashboard start [path]   # Start server (if not running) and open dashboard
kirograph dashboard stop [path]    # Stop the running engine server

dashboard start

Reads semanticEngine from .kirograph/config.json and dispatches accordingly:

• qdrant: Downloads the Qdrant Web UI on first use (cached at .kirograph/qdrant/dashboard/), spawns the Qdrant server with QDRANT__SERVICE__STATIC_CONTENT_DIR set so the dashboard is served natively, and opens http://127.0.0.1:<port>/dashboard in your browser. If the server is already running with the dashboard, reconnects instead of restarting.
• typesense: Downloads the Typesense Dashboard static UI on first use (cached at .kirograph/typesense/dashboard/), starts the Typesense server if not already running, serves the dashboard locally via a Node HTTP server, and opens it in your browser. Press Ctrl+C to stop the dashboard server — the Typesense server keeps running as a background daemon.

Both servers run as persistent daemons. The state file (.kirograph/qdrant-server.json or .kirograph/typesense-server.json) tracks the PID and port for reconnection across kg commands.

dashboard stop

Reads semanticEngine from config and sends SIGTERM to the running background process, then removes the state file. Does nothing if no server is running.

MCP Server

kirograph serve --mcp                      # Start MCP server (used by Kiro)
kirograph serve --mcp --path /my/project   # Specify project path

Configuration

KiroGraph stores its config in .kirograph/config.json. You can edit it directly.

Field	Type	Default	Description
languages	string[]	[]	Limit indexing to specific languages (empty = all)
include	string[]	[]	Glob patterns to include (empty = include everything not excluded)
exclude	string[]	see below	Glob patterns to exclude
maxFileSize	number	1048576	Skip files larger than this (bytes)
extractDocstrings	boolean	true	Extract JSDoc, docstrings, and comments
trackCallSites	boolean	true	Record line/column for call edges
enableEmbeddings	boolean	false	Generate semantic embeddings (opt-in)
embeddingModel	string	nomic-ai/nomic-embed-text-v1.5	HuggingFace feature-extraction model ID
embeddingDim	number	768	Output dimension of the chosen embedding model
semanticEngine	string	cosine	Search engine: cosine, sqlite-vec, orama, pglite, lancedb, qdrant, or typesense
useVecIndex	boolean	false	Deprecated alias for semanticEngine: "sqlite-vec"
enableArchitecture	boolean	false	Enable architecture analysis (package graph + layer detection, opt-in)
architectureLayers	object	—	Custom layer definitions: { "layerName": ["glob/**"] }
minLogLevel	string	warn	Log level: debug, info, warn, error
fuzzyResolutionThreshold	number	0.5	Name matching threshold for cross-file resolution (0.0–1.0)
cavemanMode	string	off	Agent communication style: off, lite, full, ultra

Default exclude patterns: node_modules/**, dist/**, build/**, .git/**, *.min.js, .kirograph/**

Semantic Search (Optional)

By default, KiroGraph uses exact name lookup and full-text search. Enable semantic search for natural-language queries:

{
  "enableEmbeddings": true
}

This generates vector embeddings for all functions, methods, classes, interfaces, type aliases, components, and modules using a local embedding model (downloaded automatically to ~/.kirograph/models/ on first use). Embeddings are kept in sync automatically via Kiro hooks — on every file save, create, or delete.

Run kirograph install to be guided through model and engine selection interactively with arrow-key menus, or set the fields manually in .kirograph/config.json.

Embedding models

kirograph install offers a curated selection of models compatible with @huggingface/transformers:

Model	Dim	Size	Notes
nomic-ai/nomic-embed-text-v1.5	768	~130MB	Default. Best quality for code search.
onnx-community/embeddinggemma-300m-ONNX	768	~300MB	Google Gemma-based. Multilingual, 2048-token context window.
Xenova/all-MiniLM-L6-v2	384	~23MB	Lightweight, fast. Lower accuracy.
BAAI/bge-base-en-v1.5	768	~110MB	Strong general-purpose alternative to nomic.
Custom	any	—	Any HuggingFace feature-extraction model. Provide ID + output dimension.

The embedding dimension is stored in embeddingDim in .kirograph/config.json and used to initialise all vector engines correctly. Switching models requires a full re-index (kirograph index --force).

Configure manually:

{
  "enableEmbeddings": true,
  "embeddingModel": "onnx-community/embeddinggemma-300m-ONNX",
  "embeddingDim": 768
}

Storage architecture

Each engine owns its embedding store exclusively — there is no redundant write to the main graph database:

Engine	Graph store	Vector store
cosine	kirograph.db (SQLite)	kirograph.db (vectors table)
sqlite-vec	kirograph.db (SQLite)	.kirograph/vec.db (sqlite-vec)
orama	kirograph.db (SQLite)	.kirograph/orama.json (Orama)
pglite	kirograph.db (SQLite)	.kirograph/pglite/ (PGlite+pgvector)
lancedb	kirograph.db (SQLite)	.kirograph/lancedb/ (Apache Lance)
qdrant	kirograph.db (SQLite)	.kirograph/qdrant/ (Qdrant embedded)
typesense	kirograph.db (SQLite)	.kirograph/typesense/ (Typesense embedded)

The graph store (kirograph.db) always holds nodes, edges, files, and all structural data regardless of which engine is active.

Engine comparison

Engine	Search type	Extra deps	Native?	Best for
cosine (default)	Exact cosine, linear scan	none	—	Small / medium projects, zero setup
sqlite-vec	ANN (approximate), sub-linear	better-sqlite3, sqlite-vec	yes	Large codebases, fast ANN search
orama	Hybrid (full-text + vector)	@orama/orama, @orama/plugin-data-persistence	no (pure JS)	Best result quality, no native deps
pglite	Hybrid (full-text + vector), exact	@electric-sql/pglite	no (pure WASM)	Exact results, no native deps, PostgreSQL semantics
lancedb	ANN (approximate), sub-linear	@lancedb/lancedb	no (pure JS)	Fast ANN search, no native compilation required
qdrant	ANN (HNSW), sub-linear	qdrant-local	yes (binary)	Full Qdrant feature set, HNSW index, embedded binary
typesense	ANN (HNSW), sub-linear	typesense	yes (binary)	Fast ANN search, auto-downloaded binary, no manual install

All non-cosine engines fall back silently to cosine if their optional dependencies are not installed.

cosine (default)

In-process cosine similarity over all stored embeddings. No extra dependencies. Embeddings are stored in the vectors table inside kirograph.db.

{
  "enableEmbeddings": true,
  "semanticEngine": "cosine"
}

sqlite-vec

Approximate nearest-neighbour (ANN) index stored in .kirograph/vec.db. Sub-linear search time — ideal for large codebases with thousands of indexed symbols. The SQLite vectors table is not written to; vec.db is the sole embedding store.

{
  "enableEmbeddings": true,
  "semanticEngine": "sqlite-vec"
}

npm install better-sqlite3 sqlite-vec

Requires two native dependencies (compiled C extensions). If not installed, falls back to cosine.

orama

Hybrid search powered by Orama — combines full-text relevance and vector similarity in a single query, producing higher-quality results than running the two searches separately. The index is persisted to .kirograph/orama.json and is the sole embedding store. Pure JS, no native compilation required.

{
  "enableEmbeddings": true,
  "semanticEngine": "orama"
}

npm install @orama/orama @orama/plugin-data-persistence

If not installed, falls back to cosine.

pglite

Hybrid search powered by PGlite — a WASM-compiled PostgreSQL with the pgvector extension. Combines exact nearest-neighbour vector search with full-text ranking (ts_rank) in a single SQL query. The database is persisted to .kirograph/pglite/ using PostgreSQL's WAL-based storage and is the sole embedding store. Pure WASM — no native compilation required.

{
  "enableEmbeddings": true,
  "semanticEngine": "pglite"
}

npm install @electric-sql/pglite

Key advantages:

• Exact vector results (not approximate) — deterministic and reproducible
• Native SQL ON CONFLICT upsert — no remove+insert workaround
• HNSW index (vector_cosine_ops) keeps search fast as the index grows
• Single dependency, zero native binaries

If not installed, falls back to cosine.

LanceDB

ANN vector search powered by LanceDB — stores embeddings in Apache Lance columnar format at .kirograph/lancedb/. Sub-linear search time using cosine distance. Pure JS, no native compilation required.

{
  "enableEmbeddings": true,
  "semanticEngine": "lancedb"
}

npm install @lancedb/lancedb

Key characteristics:

• Columnar storage (Apache Lance format) — efficient for batch reads and writes
• ANN cosine search — fast, sub-linear query time
• Pure JS — no native binaries or WASM required

If not installed, falls back to cosine.

qdrant

ANN vector search powered by Qdrant running in embedded mode. The engine spawns the Qdrant binary as a managed child process, persisting data to .kirograph/qdrant/. Uses @qdrant/qdrant-js as the REST client.

{
  "enableEmbeddings": true,
  "semanticEngine": "qdrant"
}

npm install qdrant-local

Key characteristics:

• HNSW index — high-quality ANN search with Qdrant's native indexing
• Embedded binary — no separate server setup; the process is spawned and managed automatically
• Persistent daemon — the server stays running between kg commands; state tracked in .kirograph/qdrant-server.json
• Built-in dashboard — run kg dashboard start to download the Qdrant Web UI and open it (cached at .kirograph/qdrant/dashboard/, served via Qdrant's built-in static content feature)
• Async startup — polls /readyz instead of blocking with a fixed sleep
• Cosine distance metric
• Data persists across restarts in .kirograph/qdrant/

Manage the server:

kirograph dashboard start   # start server + open dashboard
kirograph dashboard stop    # stop server

If not installed, falls back to cosine.

typesense

ANN vector search powered by Typesense running in embedded mode. The engine automatically downloads the Typesense server binary (~37 MB, cached at ~/.kirograph/bin/) on first use and spawns it as a managed child process. Uses the official typesense Node.js client.

{
  "enableEmbeddings": true,
  "semanticEngine": "typesense"
}

npm install typesense

Key characteristics:

• HNSW index — high-quality ANN search with Typesense's native indexing
• Auto-downloaded binary — no manual server setup; the binary is fetched and cached at ~/.kirograph/bin/ on first run
• Persistent daemon — the server stays running between kg commands; state tracked in .kirograph/typesense-server.json
• Local dashboard — run kg dashboard start to open the built-in Typesense Dashboard UI (served locally, cached at .kirograph/typesense/dashboard/)
• Async startup — polls /health instead of blocking with a fixed sleep
• Cosine distance metric
• Data persists across restarts in .kirograph/typesense/

Manage the server:

kirograph dashboard start   # start server + open dashboard
kirograph dashboard stop    # stop server

If not installed (or binary download fails), falls back to cosine.

Architecture Analysis (opt-in)

When enableArchitecture: true is set, KiroGraph analyses the high-level structure of your project during indexing and populates arch_* tables in kirograph.db. Zero behavioral change when disabled.

What it detects

Packages — logical groupings of files. Detected two ways:

1. Manifest-based — parsed from package.json, go.mod, Cargo.toml, pyproject.toml/setup.py/setup.cfg, pom.xml, build.gradle/build.gradle.kts, and .csproj files. Produces IDs like pkg:npm:src/auth.
2. Directory fallback — for files not covered by any manifest, groups them by their nearest ancestor directory. Produces IDs like pkg:dir:src/utils.

Layers — architectural tiers detected from file paths using per-language glob patterns:

Layer	Examples
api	**/controllers/**, **/routes/**, **/handlers/**, **/api/**
service	**/services/**, **/usecases/**, **/domain/**
data	**/repositories/**, **/models/**, **/db/**, **/migrations/**
ui	**/components/**, **/views/**, **/pages/**, **/screens/**
shared	**/utils/**, **/helpers/**, **/lib/**, **/common/**

Layer detection is per-language (TypeScript/JS, Python, Go, Java, Ruby, Rust, C#) with framework-specific patterns where applicable (Django, Rails, Spring MVC, ASP.NET, etc.). Custom layer overrides are supported via architectureLayers in config.

Package dependencies — rolled up from existing imports edges in the graph. No re-parsing required.

Coupling metrics — computed per package:

• Ca (afferent) — how many other packages depend on this one
• Ce (efferent) — how many packages this one depends on
• Instability — Ce / (Ca + Ce): 0 = maximally stable (everyone depends on it, it depends on nothing), 1 = maximally unstable (depends on everything, nobody depends on it)

Custom layer definitions

Override or extend the auto-detected layer patterns in .kirograph/config.json:

{
  "enableArchitecture": true,
  "architectureLayers": {
    "api": ["src/routes/**", "src/controllers/**"],
    "service": ["src/domain/**", "src/application/**"],
    "data": ["src/infrastructure/**", "src/persistence/**"]
  }
}

When architectureLayers is set, those patterns take precedence over the auto-detected ones for the specified layer names.

Storage

All architecture data is stored in kirograph.db alongside the symbol graph:

Table	Contents
arch_packages	Package definitions (id, name, path, source, language, version, deps)
arch_layers	Layer definitions (id, name, patterns)
arch_file_packages	File → package assignments
arch_file_layers	File → layer assignments (with confidence score)
arch_package_deps	Package → package dependency edges (with import count)
arch_layer_deps	Layer → layer dependency edges
arch_coupling	Per-package Ca, Ce, instability metrics

IndexProgress phase

Architecture analysis runs as a dedicated phase during kirograph index. Progress is reported with phase: 'architecture'.

Supported Languages

Language	Extensions
TypeScript	.ts
JavaScript	.js
TSX	.tsx
JSX	.jsx
Python	.py
Go	.go
Rust	.rs
Java	.java
C	.c, .h
C++	.cpp, .cc, .cxx, .hpp
C#	.cs
PHP	.php
Ruby	.rb
Swift	.swift
Kotlin	.kt
Dart	.dart
Svelte	.svelte

Framework Detection

KiroGraph automatically detects frameworks and enriches the graph with framework-specific semantics (routes, components, lifecycle methods):

JavaScript / TypeScript: React, Next.js, React Native, Svelte, SvelteKit, Express, Fastify, Koa

Python: Django, Flask, FastAPI

Ruby: Rails

Java: Spring, Spring Boot, Spring MVC

Go: generic Go resolver

Rust: generic Rust resolver

C#: ASP.NET Core

Swift: SwiftUI, UIKit, Vapor

PHP: Laravel

Detected frameworks are stored in config and used to improve symbol extraction and resolution.

Requirements

• Node.js >= 18
• Kiro IDE

License

MIT