Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/davide-desio-eleva/kirograph/llms.txt

Use this file to discover all available pages before exploring further.

KiroGraph’s core tools are registered unconditionally — they appear in every tools/list response regardless of which feature flags are enabled. The always-on core set covers the most frequent agent workflows: building task context, searching for symbols, and inspecting symbol details. Navigation tools (kirograph_status, kirograph_files, kirograph_impact) require enableNavigation: true. Call-graph traversal tools (kirograph_callers, kirograph_callees) require trackCallSites: true. Agent utility tools (kirograph_read, kirograph_retrieve, kirograph_gain, kirograph_budget) require enableAgentUtils: true.

Core Tools (always-on)

The three always-on tools — kirograph_context, kirograph_search, and kirograph_node — are registered unconditionally and appear in every tools/list response.

kirograph_context

The primary entry point for every agent task. Given a natural-language task description, it extracts symbol tokens, runs exact lookup + full-text search + vector search against the active semantic engine, resolves imports to their definitions, expands the graph to related symbols, and returns entry points, related nodes, edges, and code snippets — all in a single call. This is the only tool that invokes the vector engine on every call. For most tasks, one call to kirograph_context is enough to orient the agent without any further tool calls.
task
string
required
Natural-language description of the task, bug, or feature to investigate. Supports CamelCase, snake_case, SCREAMING_SNAKE, and dot.notation symbol tokens.
maxNodes
number
default:"20"
Maximum number of symbols to include in the response.
detail
string
default:"full"
Code verbosity level. Options: full (complete source code), signatures (signature + docstring, ~70% fewer tokens than full), summary (file locations only, no code).
projectPath
string
default:"cwd"
Absolute path to the project root. Defaults to the server’s current working directory.
How it works:
1

Token extraction

Parses the task string for CamelCase, snake_case, SCREAMING_SNAKE, and dot.notation tokens.
2

Exact + FTS + vector search

Runs exact name lookup → SQLite FTS → vector search against the active semantic engine.
3

Import resolution

Resolves imported symbols to their definitions in the graph.
4

Graph expansion

BFS-expands through related edges (calls, imports, references, type_of) up to maxNodes.
5

Response assembly

Returns entry points, related nodes, edges, code snippets, related memory (if enableMemory), related docs (if enableDocs), and security warnings (if enableSecurity).
{
  "tool": "kirograph_context",
  "arguments": {
    "task": "Add rate limiting to the authentication endpoints",
    "detail": "signatures",
    "maxNodes": 15
  }
}

Quick symbol search by name. Returns file locations and qualified names only — no source code. Ideal for confirming a symbol exists or discovering all symbols that match a partial name before drilling in with kirograph_node.
query
string
required
Symbol name or partial name (e.g. "auth", "signIn", "UserService").
mode
string
default:"name"
Search mode. name (default) uses FTS prefix match; similar uses fuzzy substring match for broader results.
kind
string
Filter by node kind. Options: function, method, class, interface, type_alias, variable, route, component.
limit
number
default:"10"
Maximum results to return (1–100).
projectPath
string
default:"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 in the common case; vector engine only as a last resort.
{
  "tool": "kirograph_search",
  "arguments": {
    "query": "authenticate",
    "kind": "function",
    "limit": 5
  }
}

kirograph_node

Get details about a specific symbol, with configurable depth — from a quick location summary up to the full source code.
symbol
string
required
Symbol name or fully-qualified name to look up.
qualified
boolean
default:"false"
When true, treats symbol as a fully-qualified name and performs an exact match, bypassing FTS. Use when you have the qualifiedName from a prior result.
detail
string
default:"summary"
Output verbosity. Options: summary (name + location + qualified name), signatures (adds signature and docstring), full (adds complete source code).
projectPath
string
default:"cwd"
Project root path.

Call-Graph Traversal Tools (require trackCallSites: true)

kirograph_callers

Find all functions and methods that call a specific symbol. Uses BFS traversal of incoming call edges in the graph database — no vector engine involved.
symbol
string
required
Symbol name to find callers for.
limit
number
default:"20"
Maximum results to return (1–100).
projectPath
string
default:"cwd"
Project root path.
Requires trackCallSites: true in your config. Call-site edges are only written to the graph during indexing when this flag is enabled.

kirograph_callees

Find all functions and methods that a specific symbol calls. Uses BFS traversal of outgoing call edges — no vector engine involved.
symbol
string
required
Symbol name to find callees for.
limit
number
default:"20"
Maximum results to return (1–100).
projectPath
string
default:"cwd"
Project root path.
{
  "tool": "kirograph_callees",
  "arguments": {
    "symbol": "UserService.create",
    "limit": 10
  }
}

kirograph_impact

Analyze what code would be affected by changing a symbol. BFS-traverses all incoming edges (call, import, reference, and others) up to the specified depth. Use this before making changes to understand blast radius.
symbol
string
required
Symbol name to analyze impact for.
depth
number
default:"2"
BFS traversal depth. Higher values find more transitive dependents.
projectPath
string
default:"cwd"
Project root path.
{
  "tool": "kirograph_impact",
  "arguments": {
    "symbol": "validateToken",
    "depth": 3
  }
}

kirograph_status

Check index health and statistics. Returns files indexed, symbol count, edge count, breakdown by kind and language, detected frameworks, database size, semantic engine status, and (if enabled) docs, data, security, and pattern subsystem summaries.
projectPath
string
default:"cwd"
Project root path.

kirograph_files

List the indexed file structure with filtering and multiple output formats. Useful for exploring the project layout and checking which files have been indexed.
filterPath
string
Filter by directory path prefix (e.g. "src/").
pattern
string
Filter by glob pattern (e.g. "**/*.ts").
maxDepth
number
Limit directory tree depth.
format
string
default:"tree"
Output format. Options: tree (visual directory tree), flat (one path per line), grouped (grouped by directory), compact (rtk-style summary with counts).
includeMetadata
boolean
default:"true"
Include language and symbol count annotations alongside each file.
projectPath
string
default:"cwd"
Project root path.

Agent Utility Tools (require enableAgentUtils: true)

kirograph_read

Read a file with session-level caching. The first read returns full content; subsequent reads of unchanged files return a compact [cached: file unchanged] marker (~13 tokens), saving context on repeated reads. Supports multiple modes for targeted extraction.
path
string
required
File path (absolute or relative to project root).
mode
string
default:"full"
Read mode. Options: full (entire file), map (structure overview), signatures (function signatures only), diff (changes since last read), lines (line range, requires start/end), imports, exports.
start
number
Start line for lines mode.
end
number
End line for lines mode.
noCache
boolean
default:"false"
Force a fresh filesystem read, bypassing the session cache.
projectPath
string
default:"cwd"
Project root path.

kirograph_retrieve

Cached Content Retrieval (CCR). Returns the full content stored in the session cache, or reads and caches the file if not yet seen. Use after kirograph_read returns a [cached: file unchanged] marker to recover the actual content without a redundant filesystem read.
path
string
required
File path to retrieve (absolute or relative to project root).
projectPath
string
default:"cwd"
Project root path.

kirograph_budget

Show current session context budget usage: tokens consumed, remaining budget, and utilization percentage.
reset
boolean
default:"false"
Reset session budget counters.
projectPath
string
default:"cwd"
Project root path.

kirograph_gain

Show token savings statistics accumulated from compressed command outputs and graph tool calls in the current session (or longer periods).
period
string
default:"session"
Time window to aggregate. Options: session (since server start), today, week, all.
projectPath
string
default:"cwd"
Project root path.

Shell Execution Tool (requires enableShellExec: true)

kirograph_exec

Run a shell command with token-optimized, noise-filtered output. Automatically applies structural filters for git, test runners (Jest, Vitest, pytest), linters (ESLint, Pylint), build tools (webpack, tsc), and Docker — stripping repeated noise lines and compacting progress bars before the output reaches the model.
command
string
required
Shell command to execute.
cwd
string
default:"project root"
Working directory for the command.
level
string
default:"normal"
Compression intensity. Options: normal (balanced), aggressive (more compact), ultra (maximum compression — may drop some detail).
timeout
number
default:"60"
Timeout in seconds before the command is killed.
projectPath
string
default:"cwd"
Project root path used for configuration lookup.
When enableShellExec is false, the tool is not registered. Use kirograph_compress instead for manual compression of command output that arrived from outside KiroGraph.
{
  "tool": "kirograph_exec",
  "arguments": {
    "command": "npm test -- --reporter=verbose",
    "level": "aggressive"
  }
}

General Compression Tool (requires enableGeneralCompression: true)

kirograph_compress

On-demand compression for arbitrary text before it reaches the model. Routes to one of two engines depending on whether command is provided. Reports inline savings: [42% tokens saved | 1800→1044 | rtk:git:aggressive].
text
string
required
Text to compress.
command
string
Shell command that produced the text (e.g. "git log", "npm test"). When provided, activates rtk-style structural filters. Omit for prose — uses caveman grammar instead.
level
string
default:"full"
Compression intensity. Options: lite / normal (light), full / aggressive (medium), ultra (maximum).
Use kirograph_compress when command output arrived outside of kirograph_exec — for example, from a tool that runs commands directly. For source code, prefer detail: "signatures" on kirograph_context instead, since compressing source code may break identifiers or whitespace-sensitive syntax.

Build docs developers (and LLMs) love