I Built a Code Intelligence Engine — Here's How CIE Works

Francisco Pérez — Fri, 23 Jan 2026 21:45:42 +0000

The Problem

AI coding assistants are everywhere — Claude Code, Cursor, GitHub Copilot. But they all share the same limitation: they don't truly understand your codebase.

They see files in isolation. Ask "how does authentication work in this project?" and they'll grep for "auth" and hope for the best.

I wanted real code intelligence.

What I Built

CIE (Code Intelligence Engine) — 20+ tools that give AI assistants deep understanding of your entire codebase.

Free, open-source, runs 100% locally.

GitHub: github.com/kraklabs/cie

The Killer Features

1. Multi-Pattern Search

Search for multiple patterns in ONE call:

Query: ["TODO", "FIXME", "HACK", "BUG"]
Result: 47 matches, grouped by pattern with file locations

This replaces running 4 separate grep commands. Game changer for batch searches and codebase audits.

2. Security Audit Automation

cie_verify_absence confirms dangerous patterns DON'T exist in your codebase:

Query: Verify no hardcoded API keys or passwords
Result: ✅ Verified — no matches found

Critical for CI/CD security pipelines. Automate what used to be manual code review.

3. Complete Call Graph Analysis

See who calls a function AND what it calls — in one view:

Query: Analyze UserService.CreateUser

Result:
Callers (3): RegisterHandler, AdminController, ImportService
Callees (12): ValidateEmail, HashPassword, db.Insert, SendWelcomeEmail...

Understand the full impact of changing any function.

4. Semantic Search That Actually Works

Find code by meaning, not just text:

Query: "validate user input before saving to database"
Result: SanitizeUserInput (87% match) — internal/validation/user.go:42

It found the right function even though the name doesn't match your query. That's the power of embeddings.

5. Instant API Discovery

cie_list_endpoints finds all your HTTP endpoints automatically:

[GET]    /api/users           → GetUsers
[POST]   /api/users           → CreateUser
[GET]    /api/users/:id       → GetUserByID
[PUT]    /api/users/:id       → UpdateUser
[DELETE] /api/users/:id       → DeleteUser

cie_list_services does the same for gRPC services.

6. Module Structure at a Glance

cie_directory_summary shows what's in each package:

internal/auth/
├── middleware.go    → AuthMiddleware, ValidateToken, RefreshToken
├── jwt.go           → GenerateJWT, ParseJWT, ValidateClaims
└── oauth.go         → OAuthCallback, ExchangeCode

Perfect for onboarding or understanding unfamiliar code.

All 20+ Tools

Search & Discovery

cie_grep — Multi-pattern literal search
cie_search_text — Regex search
cie_semantic_search — Natural language code search
cie_find_function — Direct name lookup

Architecture Analysis

cie_analyze — Answer architectural questions
cie_list_endpoints — HTTP API discovery
cie_list_services — gRPC service listing
cie_directory_summary — Module structure overview

Dependency Analysis

cie_find_callers — Who calls this function?
cie_find_callees — What does this function call?
cie_get_call_graph — Combined view of both
cie_trace_path — Trace execution from A to B

Type System

cie_find_type — Find structs/interfaces
cie_find_implementations — Find interface implementations

Security

cie_verify_absence — Verify patterns don't exist

Advanced

cie_raw_query — Custom CozoScript queries

Recommended Workflows

Exploring a new codebase:

cie_directory_summary → understand structure
cie_list_endpoints → map API surface
cie_semantic_search → find specific features

Debugging:

cie_find_function → locate the function
cie_get_call_graph → see all dependencies
cie_trace_path → understand execution flow

Security review:

cie_verify_absence → check for secrets, credentials
cie_grep → batch search for security patterns

How It Compares

Feature	Cursor	GitHub Copilot	CIE
Price	$20-200/mo	$10-39/mo	Free
Multi-pattern search	No	No	Yes
Security audit (verify_absence)	No	No	Yes
Call graph analysis	No	No	Yes
Self-hosted	No	No	Yes
Privacy	Cloud	Cloud	100% Local

Tech Stack

Go — fast, single binary
Tree-sitter — AST parsing (Go, Python, JS, TS, more)
CozoDB + RocksDB — Datalog queries for graph traversal
Ollama — local embeddings
Docker Compose — simple infrastructure

Quick Start

brew tap kraklabs/cie && brew install cie

cd /path/to/your/repo
cie init -y
cie start
cie index

Configure your MCP client (Claude Code, Cursor):

{
  "mcpServers": {
    "cie": {
      "command": "cie",
      "args": ["--mcp"]
    }
  }
}

Try It