OpenFang

The second major feature group is the comprehensive security infrastructure consisting of 16 discrete systems. These include a WASM dual-metered sandbox that enforces both fuel and epoch interruption constraints, Ed25519 manifest signing for code integrity, a Merkle audit trail for tamper-proof logging, taint tracking to prevent data leakage, SSRF protection against server-side request forgery, secret zeroization to clear sensitive data, HMAC-SHA256 mutual authentication for peer-to-peer connections, a GCRA rate limiter to control request frequency, subprocess isolation with cleared environment variables, a prompt injection scanner, and path traversal prevention. Together, these systems create a defense-in-depth architecture that protects against both external attacks and internal misconfigurations. The sandbox, in particular, ensures that tool code cannot escape its environment, making OpenFang suitable for enterprise deployments where security is non-negotiable.

The third feature group covers channel adapters and protocols, which enable agents to communicate across virtually any platform. OpenFang ships with 40 built-in channel adapters, including Telegram, Discord, Slack, WhatsApp, Microsoft Teams, IRC, and Matrix. Each adapter supports per-channel model overrides, DM and group policies, rate limiting, and custom output formatting. Beyond simple messaging, OpenFang implements three protocol layers: Model Context Protocol (MCP) as both client and server for tool interoperability, Google's Agent-to-Agent (A2A) protocol for cross-framework collaboration, and the OpenFang Protocol (OFP) for peer-to-peer networking with HMAC-SHA256 mutual authentication. This means a single agent can simultaneously serve users on Discord, send alerts to Slack, and coordinate with other agents via A2A—all without manual bridging.

OpenFang works by providing a cohesive runtime that handles every aspect of agent lifecycle. Installation is a single curl command that downloads a statically linked binary. Configuration is centralized in a `config.toml` file with 62 environment variables for fine-tuning. Spawning an agent or activating a Hand is done via the CLI. Once active, agents execute their tasks autonomously, using built-in tools (38 native plus MCP) and accessing persistent memory backed by SQLite with vector embeddings. Memory supports cross-channel canonical sessions, meaning a conversation started on Telegram can continue on Slack with full context. Automatic compaction prevents memory bloat, and JSONL session mirroring provides a full audit trail. The workflow engine allows chaining multiple agents in pipelines with fan-out, conditional, and loop steps. All activity is visible in the Tauri 2.0 desktop dashboard, which provides real-time metrics, system tray integration, and global shortcuts.

Concrete use cases demonstrate the practical impact of OpenFang. A content creator uses Clip.hand to automatically turn recorded webinars into short clips with captions and thumbnails, publishing them daily to Telegram and WhatsApp without manual editing. A sales team deploys Lead.hand to scan websites, score leads from 0 to 100, build ideal customer profiles, and export the results every morning. A security researcher uses Collector.hand for OSINT-style monitoring, detecting changes on target sites and tracking sentiment, with critical alerts sent to a dedicated channel. A professional forecaster uses Predictor.hand to collect signals, build reasoning chains, and make calibrated predictions, tracking accuracy with Brier scores. A social media manager relies on Twitter.hand to maintain a consistent posting schedule across seven content formats, with an approval queue for brand safety. These scenarios show how OpenFang transforms repetitive data tasks into automated, reliable workflows.

OpenFang is built for AI engineers, security professionals, content creators, sales teams, and open source developers who need a robust, autonomous agent platform. It runs on macOS, Linux, and Windows, with native desktop support via Tauri 2.0. The technology stack is Rust for performance and safety, WASM for sandboxed execution, SQLite for persistent storage, and a 14-crate workspace architecture. All code is released under the permissive MIT license, with a growing community of 16.8k GitHub stars. Pricing is free and open source, with installation taking under two minutes. For teams ready to move beyond manual chat-based agents, OpenFang provides the operating system for truly autonomous, secure, and multi-channel AI agents that work on your schedule.

Updated 2026-03-02