New:Socket for Asana Is Now Available.Learn more
Get Started

farai

Package Overview
Dependencies
Maintainers
1
Versions
20
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

farai

Cyber-first freestyle local AI agent

latest
Source
npmnpm
Version
0.2.9
Version published
Weekly downloads
2.8K
2137.8%
Maintainers
1
Weekly downloads
 
Created
Source

Farai

Cyber-first local AI agent for terminal-driven security work.

npm version License: Apache-2.0 GitHub stars GitHub forks

Farai terminal demo

Farai is a local AI agent built for CTFs, lab automation, security research, and long-running technical workflows. It runs from the terminal, keeps sessions resumable, executes tools locally or inside a Kali Docker runtime, and preserves tool output, artifacts, usage, and evidence so work can be reviewed instead of only trusted from a final answer.

For CSI/CyBench materials and exploratory benchmark artifacts, see farai-csi-bench.

Installation

npm install -g farai

Requirements:

  • Bun 1.1+
  • Docker
  • A model provider/API key, unless using an already configured default provider

Setup

Configure Farai and prepare the local runtime:

farai setup

Usage

Open the interactive:

farai

Config and auth files live under:

~/.local/pajarori/farai/

Farai checks the separate farai-data content channel before opening the TUI. Each commit to that repository is built into an immutable, commit-pinned bundle; when a validated bundle publishes new knowledge or skills, Farai asks before downloading it. The current content can be inspected or managed without starting a session:

farai update status
farai update check
farai update apply
farai update rollback

Set FARAI_CONTENT_MANIFEST_URL for a private channel or local file:// manifest, and FARAI_CONTENT_DIR to isolate the local content store. Set FARAI_DISABLE_CONTENT_UPDATE=1 to disable the channel.

Status

Farai is under active development.

The current defensible claim is that Farai has substantial implementation validation and an evidence-oriented agent architecture. Stronger claims require clean benchmark campaigns, stronger isolation, canonical scoring, and larger empirical evaluations.

Early benchmark artifacts should be treated as exploratory unless the run artifact, oracle, isolation policy, model selection, and challenge materials are frozen and auditable.

License

Apache License 2.0

Keywords

ai-agent

FAQs

Package last updated on 04 Sep 2026

Related posts