mkit

README

mkit: The Manglekit CLI

mkit is the command-line interface for Manglekit, the Neuro-Symbolic AI Governance framework. It provides a suite of tools to generate policies, evaluate logic, inspect data structures, and manage knowledge graphs.

Installation

go install github.com/duynguyendang/manglekit/cmd/mkit@latest

Commands

1. mkit eval

Evaluate a Datalog query against a policy blueprint, input data (JSON), and optional knowledge graph facts.

Usage:

mkit eval -p <policy.dl> -q <query> [flags]

Flags:

• -p, --policy (Required): Path to the Datalog policy file (.dl).
• -q, --query (Required): The Datalog query to execute (e.g., deny(S, R)?).
• -d, --data: Path to a JSON input file to be loaded as facts.
• -k, --knowledge (or --facts): Path to a Knowledge Graph file (.nq, .nt, .ttl).

Example:

mkit eval -p policy.dl -d input.json -q "deny(S, R)?"

---

2. mkit generate rule

Generate a Datalog rule from a natural language description using an LLM. This command uses a Neuro-Symbolic Feedback Loop (Teacher-Student Protocol) to ensure valid syntax.

Usage:

mkit generate rule --model <model_name> --prompt <description> [flags]

Flags:

• --model (Required): Model name to use (e.g., gemini-2.0-flash, gpt-4o).
• --provider: LLM Provider (google or openai). Default: google.
• --prompt (or --desc): The natural language description of the policy.
• --vocab (or --facts): Domain vocabulary (predicates) to guide generation.
• --sample: Path to a sample file (.json or .nq) to infer schema/structure.
• --icl: Path to a custom Datalog example file for In-Context Learning.

Example:

mkit generate rule \
  --model gemini-2.0-flash \
  --prompt "Deny requests if the user is not an admin and the amount > 1000" \
  --sample request.json

---

3. mkit inspect struct

Inspect a JSON structure and visualize how Manglekit converts it into Datalog facts. Useful for debugging schema mapping.

Usage:

mkit inspect struct --json <file_or_string>

Flags:

• --json (Required): Path to a JSON file or a raw JSON string.

Example:

mkit inspect struct --json request.json

---

4. mkit kg convert

Convert between different Knowledge Graph formats (e.g., Turtle to N-Quads).

Usage:

mkit kg convert -i <input_file> -o <output_file> [flags]

Flags:

• -i, --input (Required): Input file path.
• -o, --output (Required): Output file path.
• --from: Input format (auto-detected if empty).
• --to: Output format (nquads, ntriples, jsonld). Default: nquads.

Example:

mkit kg convert -i data.ttl -o data.nq

---

5. mkit serve

Start the Manglekit HTTP Server to expose the SDK via an API.

Usage:

mkit serve [flags]

Flags:

• -p, --port: Port to listen on (default: 8080).
• -f, --policy: Path to the Datalog blueprint/policy file to enforce.
• -m, --mcp: Path to an MCP configuration file.

Example:

mkit serve --port 9090 --policy firewall.dl

Environment Variables

Variable	Description
GOOGLE_API_KEY	Required for Google (Gemini) models in generate rule.
OPENAI_API_KEY	Required for OpenAI models in generate rule.